GIT repositories

Index page of all the GIT repositories that are clonable form this server via HTTPS. Übersichtsseite aller GIT-Repositories, die von diesem Server aus über git clone (HTTPS) erreichbar sind.

Services

A bunch of service scripts to convert, analyse and generate data. Ein paar Services zum Konvertieren, Analysieren und Generieren von Daten.

GNU octave web interface

A web interface for GNU Octave, which allows to run scientific calculations from netbooks, tables or smartphones. The interface provides a web form generator for Octave script parameters with pre-validation, automatic script list generation, as well presenting of output text, figures and files in a output HTML page. Ein Webinterface für GNU-Octave, mit dem wissenschaftliche Berechnungen von Netbooks, Tablets oder Smartphones aus durchgeführt werden können. Die Schnittstelle beinhaltet einen Formulargenerator für Octave-Scriptparameter, mit Einheiten und Einfabevalidierung. Textausgabe, Abbildungen und generierte Dateien werden abgefangen und in einer HTML-Seite dem Nutzer als Ergebnis zur Verfügung gestellt.

C++ OpenCV wrapper class template

C++ OpenCV wrapper class template

Template basierter Wrapper für OpenCV. Es handelt sich um eine kleine Ansammlung von Templates im namespace sw::im um die Hauptklasse cvimage (Defaultspezialiserung), deren wichtigste Instanzvariable eine cv::Mat ist, welche die Binärinformationen des Bildes enthält. Weitere Klassen dienen der eindeutigen Darstellung von geometrischen Objekten (circle, rect, rotated_rect, ellipse, contour polygon) sowie von Hilfsobjekten (z.B. histogram, color, contour_list, contour_tree). Es ist nach wie vor möglich, OpenCV Funktionen direkt auf das bild anzuwenden (mit cvimage::mat()). Da die meisten Methoden inline sind entsteht sehr wenig "Overhead" im Bezug auf die Rechenleistung.

Durch JavaDoc/Doxygen Methodendokumentation (und der expliziten Deklarationen) werden Autovervollständigungsfunktionen in Entwicklungsumgebungen besser unterstützt.

(Die weitere Dokumentation in Englisch, Source code ganz unten.)

Template based wrapper for OpenCV image processing functionality. It is a collection of templates in namespace sw::im around the main class cvimage (default specialisation), which contains a cv::Mat as main instance variable. Other class definitions are related to nonambiguous definitions of geometry objects (e.g. circle, rect, rotated_rect, ellipse, contour, polygon) and Auxiliary objects (e.g. histogram, color, contour_list, contour_tree). Using the inlined "cvimage::mat()" method it is possible to work with the matrix as usually done in OpenCV. As most of the methods are inlined the computational overhead is minimal.

JavaDoc/Doxygen style method documentation allows smart autocomplete in IDEs.

Features

Briefely shown, here the methods that the class provides:

// I/O, methods known from STL containers:
img.load(filename, [as-is, gray, color]); // Load
img.save([filename]);                  // Save (file omitted===load name)
img.clear();                           // Reset image, clear matrix
if(img.empty()) ...                    // Empty image?
 
// Basic information
w = img.width();
h = img.height();
A = img.area();
n = img.channels();                    // Number of channels
if(img.is_rgb()) {}
if(img.is_gray()) {}
if(img.is_hsv()) {}
 
// Information requiring calculations:
n = img.num_nonzero_pixels();          // Count pixels that are not black
rct = img.bounding_rect();             // Gradient based border removal
my_color = img.min();                  // Minimum values of all channels
my_color = img.max();                  // Maximum values of all channels
my_color = img.mean();                 // Average values of all channels
my_color = img.color_at(x,y);          // Colour of pixel x/y
 
// Color conversion, channel selection:
img.to_grayscale();
img.to_rgb();
img.to_hsv();
img2 = img.channel_red();              // Extract channel to a new gray
img2 = img.channel_blue();             // scale image ...
img2 = img.channel_green();
img2 = img.channel_hsv_hue()
img2 = img.channel_hsv_sat();
img2 = img.channel_hsv_val();
imgs = img.split_channels();           // All channels as vector of images
imgs = img.split_channels_hsv();       // All channels as HSV
 
// Methods returning new images:
img2 = img.subimage(rect2d(x0,y0,x1,y1));
img2 = img.in_color_range(<color>,[max_difference]);
 
// Methods modifying the image and returning a reference to itself:
img.negate();                          // Invert colors
img.extend(<where?>, <how?>, <color>); // Extend the image area, make border
img.crop(<where (rectangle)>);         // Crop the image
img.resample_up();                     // Make it double the image size
img.resample_down();                   // Make it half the image size
img.resize_stretch(<width>,<height>,[interpolation]); // Resize, stretch
img.resize_scale(<factor>);            // .. so that the area is factor..
img.resize(<width>,<height>);          // To min width/height, keep aspect ratio
img.dilate([radius]);                  // Diliate image
img.erode([radius]);                   // Erode image
img.blur([radius], [how?]);            // Blur with selected method
img.sobel(...);                        // Sobel filter
img.scharr(...);                       // Scharr filter
img.laplacian(...);                    // Laplace derivative
img.gradient(...);                     // Gradient (uses sobel() in x/y)
img.edges_canny(...);                  // Canny edge detection
img.corners_harris(...);               // Harris corner detector
img.level([black-threshold],[white-threshold]); // Colour range --> 0..MAX
img.histogram_equalize();              // Similar effect as level()
img.normalize();                       // Normalise image
img.threshold([tsh]);                  // All pixels <tsh=0, all >=tsh =MAX
img.threshold_adaptive([radius]);      // Adaptive threshold
 
// Contours, feature detection:
cntrs = img.contours(...);             // Find contours, return them as plain list
cntrs = img.contours_ccomp(...);       // Return them as externals / holes
cntrs = img.contours_tree(...);        // Return them as tree object
lines = img.hough_lines(...);          // Hough line transform
lines = img.hough_circles(...);        // Hough circle transform
hist  = img.histogram(...);            // Calculate histogram
 
// Drawing
img.draw( <any geometry object> );     // Draw (contours) of ...
img.fill( <any geometry object> );     // Fill area of ... with color ...

Usage annotations:

Dateien

Files

cvimage.hh microtest/cvimage.cc

Beispiele

Examples

/**
 * @ccflags -Wall -O3 -pedantic
 * @ldflags -lm -lopencv_core -lopencv_highgui -lopencv_imgproc
 */
#include <sw/cvimage.hh>
#include <cmath>
#include <vector>
#include <iostream>
#include <sstream>
#define DBG(_X) cerr << _X << endl
 
using namespace std;
using namespace cv;
using namespace sw;
using namespace sw::im;
 
/**
 * Main example / test
 * @param int argc
 * @param char** argv
 * @return int
 */
int main(int argc, char** argv)
{
  int tst = (argc>=4) ? ::atol(argv[3]) : 0;
 
  if(argc < 3) {
    cerr << "Usage: " << argv[0] << " <input image> <output image>" << endl;
    exit(1);
  }
  cvimage im;
  im.load(argv[1]);
 
  //////////////////////////////////////////////////////////////////////////////
 
  if(tst--==0) {
    DBG("load, save identical images)");
  } else if(tst--==0) {
    DBG("Contours (ccomp)");
    im.resize(500,500);
    cvimage im2 = im;
    im2.to_grayscale().gradient().level(0.1,1).threshold(0.2).dilate(3);
    cvimage::contour_ccomp_list_t cn = im2.contours_ccomp();
    im.to_rgb();
    for(unsigned i=0; i<cn.outer.size(); i++) {
      im.draw(cn.outer[i], "#00ff00");
    }
    for(unsigned i=0; i<cn.inner.size(); i++) {
      im.draw(cn.inner[i], "#ff0000");
    }
  } else if(tst--==0) {
    DBG("Resize to max width/height");
    im.resize(500,500);
  } else if(tst--==0) {
    DBG("Image.draw(image)");
    cvimage iim = im;
    int w=im.width(), h=im.height(), c=im.width()/4;
    cvimage::rect_t r(c, c, c+w/2, c+h/2);
    im.draw(iim, r);
    im.draw(r, "#ff0088", 2);
  } else if(tst--==0) {
    DBG("Extend");
    int w=im.width(), h=im.height();
    im.resize_scale(0.5).extend(w, h, false, cvimage::em_color, cvimage::color_t(10));
  } else if(tst--==0) {
    DBG("Resize");
    im.resize_stretch(200,200);
  } else if(tst--==0) {
    DBG("Crop");
    im.crop(im.width()*0.5, im.height() * 0.5);
    im.crop(im.width()*1.2, im.height() * 0.5);
    im.crop(im.width()*1.2, im.height() * 1.5);
    im.crop(5, 5);
  } else if(tst--==0) {
    DBG("Mean colour");
    DBG("Color is: " << im.color_mean());
  } else if(tst--==0) {
    DBG("Adaptive threshold");
    im.threshold_adaptive();
  } else if(tst--==0) {
    DBG("Resize");
    im.resize_scale(0.1);
  } else if(tst--==0) {
    DBG("Histogram");
    cvimage::histogram_t hist = im.histogram();
    int sz = 5;
    im.to_rgb().draw(hist,
      cvimage::rect_t(
        cvimage::point_t(0,(im.height())*(sz-1)/sz),
        im.width(), (im.height())/sz-1
      )
    );
    DBG("histogram=" << hist);
  } else if(tst--==0) {
    DBG("Image info");
    DBG("image=" << im);
  } else if(tst--==0) {
    DBG("Equalise histogram");
    im.histogram_equalize();
  } else if(tst--==0) {
    DBG("Floodfill color of seed point");
    for(int x=10; x<im.width(); x+= im.width()/10) {
      for(int y=10; y<im.height(); y+= im.height()/10) {
        im.floodfill(im::point2d(x,y), 4);
      }
    }
  } else if(tst--==0) {
    DBG("Color at");
    for(int x=10; x<im.width(); x+= im.width()/10) {
      for(int y=10; y<im.height(); y+= im.height()/10) {
        im::color c = im.color_at(x,y);
        DBG("[" << x << "/" << y << "]: " << c);
        im.fill(rect2d(x-10,y-10,x+10,y+10), c);
      }
    }
  } else if(tst--==0) {
    DBG("Structured contours");
    cvimage im2 = im;
    im2.to_grayscale().threshold(0.5);
    cvimage::contour_nested_list_t cnl = im2.contours_tree();
    im.to_rgb();
    for(unsigned i=0; i<cnl.children().size(); i++) {
      im.draw(cnl.children()[i], "#ff0000");
      for(unsigned j=0; j<cnl.children()[i].children().size(); j++) {
        im.draw(cnl.children()[i].children()[j], "#00ff00");
        for(unsigned k=0; k<cnl.children()[i].children()[j].children().size(); k++) {
          im.draw(cnl.children()[i].children()[j].children()[k], "#ffff00");
          for(unsigned l=0; l<cnl.children()[i].children()[j].children()[k].children().size(); l++) {
            im.draw(cnl.children()[i].children()[j].children()[k].children()[l], "#ff00ff");
          }
        }
      }
    }
  } else if(tst--==0) {
    DBG("Floodfill mask");
    im = im.floodfill_mask(im::point2d(0,0), 8);
  } else if(tst--==0) {
    DBG("Floodfill");
    im.floodfill(im::point2d(0,0), "#ff0000", 8);
  } else if(tst--==0) {
    DBG("Color range selection");
    im = im.in_color_range("#888888", 0.5, true);
  } else if(tst--==0) {
    DBG("Bounding rect / sub image");
    cvimage::rect_t r = im.bounding_rect();
    DBG("width  = " <<  im.width());
    DBG("height = " <<  im.height());
    DBG("Rect=" << r);
    im.to_rgb();
    im.draw(r, "#ff0000");
    im = im.subimage(r);
  } else if(tst--==0) {
    DBG("Mean");
    DBG("Mean color is " << im.mean() );
  } else if(tst--==0) {
    DBG("Nonzero pixels");
    unsigned nz = im.threshold(0.8).num_nonzero_pixels();
    DBG("width  = " <<  im.width());
    DBG("height = " <<  im.height());
    DBG("area   = " <<  im.area());
    DBG("nonzero= " <<  nz);
    DBG(nz << " = " << (100.0 * nz/im.area()) << "%");
  } else if(tst--==0) {
    DBG("Resample down");
    im.resample_down();
  } else if(tst--==0) {
    DBG("HSV H channel");
    im = im.channel_hsv_hue();
  } else if(tst--==0) {
    DBG("HSV S channel");
    im = im.channel_hsv_sat();
  } else if(tst--==0) {
    DBG("HSV V channel");
    im = im.channel_hsv_val();
  } else if(tst--==0) {
    DBG("Red channel");
    im = im.channel_red();
  } else if(tst--==0) {
    DBG("Green channel");
    im = im.channel_green();
  } else if(tst--==0) {
    DBG("Blue channel");
    im = im.channel_blue();
  } else if(tst--==0) {
    DBG("Canny edges");
    im = im.edges_canny();
  } else if(tst--==0) {
    DBG("Harris corners");
    im = im.corners_harris();
  } else if(tst--==0) {
    DBG("Gradient");
    im = im.to_grayscale().gradient();
  } else if(tst--==0) {
    DBG("Sharr x");
    im = im.to_grayscale().scharr(1,0);
  } else if(tst--==0) {
    DBG("Sharr y");
    im = im.to_grayscale().scharr(0,1);
  } else if(tst--==0) {
    DBG("Sobel x/y");
    im = im.to_grayscale().sobel(1,1);
  } else if(tst--==0) {
    DBG("Sobel x");
    im = im.to_grayscale().sobel(1,0);
  } else if(tst--==0) {
    DBG("Sobel y");
    im = im.to_grayscale().sobel(0,1);
  } else if(tst--==0) {
    DBG("Laplacian");
    im = im.to_grayscale().laplacian();
  } else if(tst--==0) {
    DBG("Heatmap colors");
    for(double d=0; d<im.width(); d+=1) {
      im.fill(rect2d(point2d(d, 0), 2, 50), color::heatmap(d/im.width()));
    }
  } else if(tst--==0) {
    DBG("Contours");
    cvimage::contour_list_t contours = cvimage(im)
      .threshold(0.8)
      .negate()
      .contours()
      ;
    cerr << contours.size() << endl;
    for(unsigned i=0; i<contours.size(); i++) {
      im.draw(contours[i], "#ff0000");
    }
  } else if(tst--==0) {
 
    DBG("Draw/fill");
    contour2d ply = point2d(10,10);
    ply += point2d(1000, 10);
    ply += point2d(500, 500);
    ply += point2d(10 ,2000);
    im.fill(ply, "#ffffaa");
    im.draw(ply, "#ff0000");
 
    im.fill(rotated_rect2d(500,600,300,200, -20), "#ff88ff");
    im.draw(rotated_rect2d(500,600,300,200, -20), "#ff00ff", 10);
 
    im.fill(rect2d(900,200,100,200), "#ffffaa");
    im.draw(rect2d(900,200,100,200), "#555500", 2);
 
    im.draw(circle2d(500,500,200), "#00ffff", 10);
    im.fill(circle2d(500,500,200), "#bbffff");
 
    im.draw(ellipse2d(500,700,200,300, 45), "#555555", 10);
    im.fill(ellipse2d(500,700,200,300, 45), "#999999");
 
    im.draw(line2d(0,0, 100,500), "#0f0f0f", 4);
    im.draw(point2d(200, 200), "#0000ff", 50);
 
  } else if(tst--==0) {
 
    DBG("Hough lines");
    cvimage::line_list_t lines;
    cvimage im2 = im;
    im2.blur(2).edges_canny(100,120,3);
    lines = im2.hough_lines(1, 100, 1, 1, 1 );
    for(cvimage::line_list_t::const_iterator it=lines.begin(); it!=lines.end(); it++) {
      im.draw(*it, "#ff0000",1);
    }
 
  } else if(tst--==0) {
 
    DBG("Hough circles");
    cvimage::circle_list_t circles;
    cvimage im2 = im;
    im2.to_grayscale().blur(2);
    circles = im2.hough_circles();
    for(cvimage::circle_list_t::const_iterator it=circles.begin(); it!=circles.end(); it++) {
      im.draw(*it, "#ff0000",1);
    }
  } else {
    DBG("No test done ...");
  }
 
  im.save(argv[2]);
  return 0;
}

Quelltext

Source code

/**
 * @package de.atwillys.cc.swl.im
 * @license BSD (simplified)
 * @author Stefan Wilhelm (stfwi)
 *
 * @file cvimage.hh
 * @ccflags
 * @ldflags -lm -lopencv_core -lopencv_highgui -lopencv_imgproc
 * @platform linux, bsd, windows
 * @standard >= c++98
 *
 * -----------------------------------------------------------------------------
 *
 * Template based wrapper for OpenCV image processing functionality. Brief
 * overview (main image class is `cvimage`), example instance `img`. Most of
 * the methods have multiple overloads that are not shown here. The detailed
 * documentation about each method is in JavaDoc/Doxygen format in the source
 * code:
 *
 * - I/O, methods known from STL containers:
 *
 *  - img.load(filename, [as-is, gray, color]); // Load
 *  - img.save([filename]);                  // Save (file omitted===load name)
 *  - img.clear();                           // Reset image, clear matrix
 *  - if(img.empty()) ...                    // Empty image?
 *
 * - Basic information
 *
 *   - w = img.width();
 *   - h = img.height();
 *   - A = img.area();
 *   - n = img.channels();                    // Number of channels
 *   - if(img.is_rgb()) {}
 *   - if(img.is_gray()) {}
 *   - if(img.is_hsv()) {}
 *
 * - Information requiring calculations:
 *
 *   - n = img.num_nonzero_pixels();          // Count pixels that are not black
 *   - rct = img.bounding_rect();             // Gradient based border removal
 *   - my_color = img.min();                  // Minimum values of all channels
 *   - my_color = img.max();                  // Maximum values of all channels
 *   - my_color = img.mean();                 // Average values of all channels
 *   - my_color = img.color_at(x,y);          // Colour of pixel x/y
 *
 * - Color conversion, channel selection:
 *
 *   - img.to_grayscale();
 *   - img.to_rgb();
 *   - img.to_hsv();
 *
 *   - img2 = img.channel_red();              // Extract channel to a new gray
 *   - img2 = img.channel_blue();             // scale image ...
 *   - img2 = img.channel_green();
 *   - img2 = img.channel_hsv_hue()
 *   - img2 = img.channel_hsv_sat();
 *   - img2 = img.channel_hsv_val();
 *
 *   - imgs = img.split_channels();           // All channels as vector of images
 *   - imgs = img.split_channels_hsv();       // All channels as HSV
 *
 * - Methods returning new images:
 *
 *   - img2 = img.subimage(rect2d(x0,y0,x1,y1));
 *   - img2 = img.in_color_range(<color>,[max_difference]);
 *
 * - Methods modifying the image and returning a reference to itself:
 *
 *   - img.negate();                          // Invert colors
 *   - img.extend(<where?>, <how?>, <color>); // Extend the image area, make border
 *   - img.crop(<where (rectangle)>);         // Crop the image
 *
 *   - img.resample_up();                     // Make it double the image size
 *   - img.resample_down();                   // Make it half the image size
 *   - img.resize_stretch(<width>,<height>,[interpolation]); // Resize, stretch
 *   - img.resize_scale(<factor>);            // .. so that the edges are factor..
 *   - img.resize(<width>,<height>);   // To min width/height, keep aspect ratio
 *
 *   - img.dilate([radius]);                  // Diliate image
 *   - img.erode([radius]);                   // Erode image
 *   - img.blur([radius], [how?]);            // Blur with selected method
 *   - img.sobel(...);                        // Sobel filter
 *   - img.scharr(...);                       // Scharr filter
 *   - img.laplacian(...);                    // Laplace derivative
 *   - img.gradient(...);                     // Gradient (uses sobel() in x/y)
 *   - img.edges_canny(...);                  // Canny edge detection
 *   - img.corners_harris(...);               // Harris corner detector
 *
 *   - img.level([black-threshold],[white-threshold]); // Colour range --> 0..MAX
 *   - img.histogram_equalize();              // Similar effect as level()
 *   - img.normalize();                       // Normalise image
 *   - img.threshold([tsh]);                  // All pixels <tsh=0, all >=tsh =MAX
 *   - img.threshold_adaptive([radius]);      // Adaptive threshold
 *
 * - Contours, feature detection:
 *
 *   - cntrs = img.contours(...);       // Find contours, return them as plain list
 *   - cntrs = img.contours_ccomp(...); // Return them as externals / holes
 *   - cntrs = img.contours_tree(...);  // Return them as tree object
 *
 *   - lines = img.hough_lines(...);          // Hough line transform
 *   - lines = img.hough_circles(...);        // Hough circle transform
 *   - hist  = img.histogram(...);            // Calculate histogram
 *
 * - Drawing
 *
 *   - img.draw( <any geometry object> );     // Draw (contours) of ...
 *   - img.fill( <any geometry object> );     // Fill area of ... with color ...
 *
 * -----------------------------------------------------------------------------
 *
 * The class templates represent own data types that can be easily extended and
 * provide converters and accessors to OpenCV types, such as cv::Mat, cv::Scalar,
 * etc. The class templates are located in namespace `sw::im`, are build around
 * the "main class" ...
 *
 *  - Template  class        : template <...> class sw::im::basic_image;
 *
 *  - Default specialisation : class sw::im::cvimage;
 *
 * ... , can have the template typename arguments ...
 *
 *  - crd_t  : Typename for coordinates, default `int`
 *  - val_t  : Typename for the color representation, default `unsigned char`
 *  - float_t: Typename for used floating point values
 *  - char_t : Typename for characters, from which strings are represented
 *
 * ... and encompass the classes:
 *
 *  - template <typename crd_t, typename val_t, typename float_t> class basic_image;
 *  - template <typename char_type> class basic_image_exception;
 *  - template <typename val_t, typename float_t, val_t max_quantum> class basic_color;
 *  - template <typename crd_t, typename float_t> class basic_point2d;
 *  - template <typename crd_t, typename float_t> class basic_line2d;
 *  - template <typename crd_t, typename float_t> class basic_rect2d;
 *  - template <typename crd_t, typename float_t> class basic_circle2d;
 *  - template <typename crd_t, typename float_t> class basic_ellipse2d;
 *  - template <typename crd_t, typename float_t> class basic_rotated_rect2d;
 *  - template <typename crd_t, typename float_t> class basic_polynom2d;
 *  - template <typename crd_t, typename float_t> class basic_contour2d;
 *  - template <typename crd_t, typename val_t, typename float_t> class basic_contour_tree;
 *  - template <typename crd_t, typename val_t, typename float_t> class basic_histogram;
 *
 * The default specialisations are derived from the default specialisation of
 * `basic_image`, which holds the fitting (template-) types for its colors,
 * geometries etc:
 *
 *  - typedef detail::basic_image<int, unsigned char, double> cvimage;
 *  - typedef cvimage::exception image_exception;
 *  - typedef cvimage::color_t color;
 *  - typedef cvimage::point_t point2d;
 *  - typedef cvimage::line_t line2d;
 *  - typedef cvimage::line_list_t lines2d;
 *  - typedef cvimage::rect_t rect2d;
 *  - typedef cvimage::circle_t circle2d;
 *  - typedef cvimage::circle_list_t circles2d;
 *  - typedef cvimage::ellipse_t ellipse2d;
 *  - typedef cvimage::rotated_rect_t rotated_rect2d;
 *  - typedef cvimage::polynom_t polynom2d;
 *  - typedef cvimage::contour_t contour2d;
 *  - typedef cvimage::contour_list_t contours2d;
 *  - typedef cvimage::contour_nested_list_t nested_contours2d;
 *
 * -----------------------------------------------------------------------------
 *
 * Usage annotations:
 *
 *  - Link against: -lopencv_core -lopencv_highgui -lopencv_imgproc
 *
 *  - `cvimage`/`basic_image` work with copies when assigning (`image2=image1;`),
 *    not with references as cv::Mat.
 *
 *  - All methods that modify the image return a reference to itself, so that
 *    the methods can be chain-called, e.g.:
 *
 *      `image1.load("file.png", cvimage::load_gray).gradient().level(0.1,1)
 *             .threshold(0.2).dilate(3).erode(1).save();` , etc etc.
 *
 *  - OpenCV matrices can be accessed and used as usual using the method
 *    `cvimavge::mat()`.
 *
 *  - `cvimage`/`basic_image` provides drawing functions for each geometry class,
 *    as well as for the histogram (so that histograms can be drawn into the image):
 *
 *    - cvimage.draw(point2d/circle2d/ellipse2d ..);
 *    - cvimage.fill(point2d/circle2d/ellipse2d ..);
 *
 *  - The colour class provides reading a colour from hex string, and the compiler
 *    implicitly converts strings to colours for method arguments:
 *    - color light_gray("#aaaaaa");             // Define a light gray color
 *    - image.fill(circle2d(x,y,r), "#ff0000");  // Draw a red filled circle
 *
 *  - Colors images (if not converted to HSV) and class `color` represent the
 *    color as RGB. Because OpenCV normally uses BGR, the "internal" CV matrices
 *    are BGR. The implicit conversion operator from `color` to `cv::Scalar`
 *    returns a BGR(A) scalar as well, and the index accessors (operator[]) of
 *    color also arranges BGR, so mycolor[0] is the blue component. Normally the
 *    inline getters `r()`, `g()`, `b()` are used to access colour components.
 *
 *  - The color space used is 8 bit, t.m. CV_8UC1 for gryscale, CV_8UC3 for RGB
 *    and HSV. If you use CV functions that convert to another matrix element
 *    data type, keep in mind that `cvimage` methods might convert it back to
 *    CV_8U.
 *
 *  - Geometry objects provide short `inline` getters/setters, type converters
 *    from/to corresponding CV data types (e.g. cv::Vec2f etc), as well as the
 *    methods `area()` and `circumference()`.
 *
 *  - Image modification methods that deal with matrix element values (pixel
 *    colours) normally take `float_t` arguments and assume that the argument
 *    you pass is a float value between 0 and 1. Hence, you do not need to care
 *    about the colour quantum range (which can be defined as `color` template
 *    value argument and is normally 0 to 255). E.g.
 *
 *     - DON'T say: img.threshold(25);  // WRONG: Threshold at 100%, as 25 >= 1
 *     - but   say: img.threshold(0.1); // RIGHT; Threshold at 10%
 *
 *
 * -----------------------------------------------------------------------------
 * +++ BSD license header +++
 * Copyright (c) 2009-2014, Stefan Wilhelm (stfwi, <cerbero s@atwilly s.de>)
 * All rights reserved.
 * Redistribution and use in source and binary forms, with or without modification,
 * are permitted provided that the following conditions are met: (1) Redistributions
 * of source code must retain the above copyright notice, this list of conditions
 * and the following disclaimer. (2) Redistributions in binary form must reproduce
 * the above copyright notice, this list of conditions and the following disclaimer
 * in the documentation and/or other materials provided with the distribution.
 * (3) Neither the name of atwillys.de nor the names of its contributors may be
 * used to endorse or promote products derived from this software without specific
 * prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS
 * AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING,
 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER
 * OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
 * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
 * DAMAGE.
 * -----------------------------------------------------------------------------
 */
#ifndef SW_CVIMG_HH
#define SW_CVIMG_HH
 
// <editor-fold desc="preprocessor" defaultstate="collapsed">
#include <opencv2/core/core.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <string>
#include <exception>
#include <vector>
#include <cmath>
#include <algorithm>
#include <iostream>
#include <limits>
#include <cstdio>
#include <cassert>
#define dbg(X) std::cerr << X << std::endl;
#define dbg2(X) std::cerr << X << std::endl;
#define iex(S) throw exception(S) /* macro undefined at EOF */
// </editor-fold>
 
namespace sw { namespace im { namespace detail {
 
// <editor-fold desc="forward decls" defaultstate="collapsed">
 
template <typename crd_t, typename float_t> class basic_point2d;
template <typename crd_t, typename float_t> class basic_line2d;
template <typename crd_t, typename float_t> class basic_rect2d;
template <typename crd_t, typename float_t> class basic_circle2d;
template <typename crd_t, typename float_t> class basic_ellipse2d;
template <typename crd_t, typename float_t> class basic_rotated_rect2d;
template <typename crd_t, typename float_t> class basic_polynom2d;
template <typename crd_t, typename float_t> class basic_contour2d;
template <typename crd_t, typename val_t, typename float_t> class basic_contour_tree;
 
template <typename val_t=unsigned char, typename float_t=double, val_t max_quantum=255>
class basic_color;
 
template <typename char_type> class basic_image_exception;
template <typename crd_t, typename val_t, typename float_t> class basic_image;
template <typename crd_t, typename val_t, typename float_t> class basic_histogram;
 
// </editor-fold>
 
// <editor-fold desc="basic_image_exception" defaultstate="collapsed">
/**
 * Class basic_image_exception
 */
template <typename char_type>
class basic_image_exception : public std::exception
{
public:
 
  basic_image_exception() throw()
  { what_ = "<Unspecified>"; }
 
  explicit basic_image_exception(const char_type *wht) throw()
  { what_ = wht ? wht : "<!NULL!>"; }
 
  virtual ~basic_image_exception() throw()
  { ; }
 
  inline const char_type * what() const throw()
  { return what_; }
 
protected:
  const char_type *what_;
};
// </editor-fold>
 
// <editor-fold desc="basic_color" defaultstate="collapsed">
/**
 * Color class template
 */
template <typename val_t, typename float_t, val_t max_quantum>
class basic_color
{
public:
 
  /**
   * Default color: Black: RGB/HSV=0, alpha=0
   */
  inline basic_color() throw()
  { c_[0]=c_[1]=c_[2]=c_[3]=0; }
 
  /**
   * Gray value color: RGB/HSV=value, alpha=0
   */
  inline basic_color(val_t v) throw()
  { c_[0]=c_[1]=c_[2]=v; c_[3]=0; }
 
  /**
   * RGB color: Default alpha=0
   * @param hex
   */
  inline basic_color(val_t r, val_t g, val_t b, val_t a=0) throw()
  { c_[2]=r; c_[1]=g; c_[0]=b; c_[3]=a; }
 
  /**
   * Hex defined RGB color, alpha=255
   * @param std::string hex
   */
  inline basic_color(std::string hex)
  { operator=(hex); }
 
  /**
   * Hex defined RGB color, alpha=255
   * @param const char *hex
   */
  inline basic_color(const char *hex)
  { operator=(hex); }
 
  /**
   * cv::Scalar defined color.
   * @param cv::Scalar s
   */
  inline basic_color(cv::Scalar s) throw()
  { c_[2]=s[0]; c_[1]=s[1]; c_[0]=s[2]; c_[3]=s[3]; }
 
  /**
   * d'tor
   */
  virtual ~basic_color()
  { ; }
 
  /**
   * Assign gray scale, alpha=255
   * @param val_t c
   * @return basic_color&
   */
  inline basic_color & operator=(val_t c) throw()
  { c_[0]=c_[1]=c_[2]=c; c_[3]=0; return *this; }
 
  /**
   * Assign copy color
   * @param const basic_color &c
   * @return basic_color&
   */
  inline basic_color & operator=(const basic_color &c) throw()
  { c_[0]=c.c_[0]; c_[1]=c.c_[1]; c_[2]=c.c_[2]; c_[3]=c.c_[3]; return *this; }
 
  /**
   * Assign hex RGB, alpha=255
   * @param const char *hex
   * @return basic_color&
   */
  inline basic_color & operator=(const char *hex)
  { return operator=(std::string(hex)); }
 
  /**
   * RGB color hex notation e.g. "#ff0000" for red.
   * @param std::string hex
   * @return basic_color&
   */
  inline basic_color & operator=(std::string hex)
  {
    const char ct[] = "0123456789abcdef";
    c_[0]=c_[1]=c_[2]=c_[3]=0;
    #define h2d(c) ((int)( \
        ((c>='a' && c<='f') ? (c-'a'+10) : \
        ((c>='A' && c<='F') ? (c-'A'+10) : \
        ((c>='0' && c<='9') ? (c-'0') : \
        ((0)))))))
    if(hex.find_first_of("#") == 0) hex.erase(0, 1);
    if(hex.find_first_not_of(ct) != std::string::npos) return *this;
 
    if(hex.length() >= 6) {
      c_[0] = (h2d(hex[4])<<4) + (h2d(hex[5])<<0);  // B
      c_[1] = (h2d(hex[2])<<4) + (h2d(hex[3])<<0);  // G
      c_[2] = (h2d(hex[0])<<4) + (h2d(hex[1])<<0);  // R
      if(hex.length() < 8) return *this;
      c_[3] = (h2d(hex[6])<<4) + (h2d(hex[7])<<0);  // A
    } else if(hex.length() == 3) {
      c_[0] = (h2d(hex[2])<<4) + (h2d(hex[2])<<0);  // B
      c_[1] = (h2d(hex[1])<<4) + (h2d(hex[1])<<0);  // G
      c_[2] = (h2d(hex[0])<<4) + (h2d(hex[0])<<0);  // R
    }
    return *this;
    #undef h2d
  }
 
  /**
   * Returns true if the color is black (all values 0, alpha channel ignored)
   * @return bool
   */
  inline bool operator!() const throw()
  { return c_[0]==0 && c_[1]==0 && c_[2]==0; }
 
  /**
   * Returns true if all components of the colour are equal to the given value
   * (gray value compare). Compare ignores alpha channel.
   * @return bool
   */
  inline bool operator==(val_t c) const throw()
  { return c_[0]==c && c_[1]==c && c_[2]==c; }
 
  /**
   * Returns true if the colours are in all channels identical. Compare ignores alpha channel.
   * @return bool
   */
  inline bool operator==(const basic_color &c) const throw()
  { return c_[0]==c.c_[0] && c_[1]==c.c_[1] && c_[2]==c.c_[2]; }
 
  /**
   * Returns true if the colour identical to the given hex value.
   * @return bool
   */
  inline bool operator==(const char *hex) const
  { return *this == basic_color(std::string(hex)); }
 
  /**
   * Returns true if the colour identical to the given hex value. Compare ignores alpha channel.
   * @return bool
   */
  inline bool operator==(std::string hex) const
  { basic_color c(hex); return c_[0]==c.c_[0] && c_[1]==c.c_[1] && c_[2]==c.c_[2]; }
 
  /**
   * Returns true if all components of the colour are equal to the given value
   * (gray value compare).
   * @return bool
   */
  inline bool operator!=(val_t c) const throw()
  { return !(*this == c); }
 
  /**
   * Returns true if the colours are in all channels identical.
   * @return bool
   */
  inline bool operator!=(const basic_color &c) const throw()
  { return !(*this == c); }
 
  /**
   * Returns true if the colour identical to the given hex value.
   * @return bool
   */
  inline bool operator!=(std::string hex) const
  { return !(*this == basic_color(hex)); }
 
  /**
   * Returns true if the colour identical to the given hex value.
   * @return bool
   */
  inline bool operator!=(const char *hex) const
  { return !(*this == basic_color(std::string(hex))); }
 
  /**
   * Blue channel
   * @return val_t
   */
  inline val_t b() const throw()
  { return c_[0]; }
 
  /**
   * Green channel
   * @return val_t
   */
  inline val_t g() const throw()
  { return c_[1]; }
 
  /**
   * Red channel
   * @return val_t
   */
  inline val_t r() const throw()
  { return c_[2]; }
 
  /**
   * ALpha channel
   * @return val_t
   */
  inline val_t a() const throw()
  { return c_[3]; }
 
  /**
   * Blue channel
   * @param val_t
   */
  inline void b(val_t v) throw()
  { c_[0] = v; }
 
  /**
   * Green channel
   * @param val_t
   */
  inline void g(val_t v) throw()
  { c_[1] = v; }
 
  /**
   * Red channel
   * @param val_t
   */
  inline void r(val_t v) throw()
  { c_[2] = v; }
 
  /**
   * ALpha channel
   * @param val_t
   */
  inline void a(val_t v) throw()
  { c_[3] = v; }
 
  /**
   * Note: This is CV BGR format. color[0] is blue not red.
   * @param unsigned i
   * @return val_t
   */
  inline val_t operator[](unsigned i) const
  { return c_[i]; }
 
  /**
   * Note: This is CV BGR format. color[0] is blue not red.
   * @param unsigned i
   * @return val_t
   */
  inline val_t& operator[](unsigned i)
  { return c_[i]; }
 
  /**
   * To cv scalar
   * @return cv::Scalar
   */
  inline operator cv::Scalar () const throw()
  { return cv::Scalar(c_[0], c_[1], c_[2], c_[3]); }
 
  /**
   * To std::string
   * @return std::string
   */
  inline operator std::string () const
  {
    char c[16];
    ::snprintf(c,16,"#%02x%02x%02x", (int)c_[2]&0xff,(int)c_[1]&0xff,(int)c_[0]&0xff);
    std::string s(c);
    if(c_[3]) { ::snprintf(c, 16,"%02x", (int)c_[3]&0xff); }
    return s;
  }
 
  /**
   * Returns the maximum quantum value (for 8bit === 255 === 0xff)
   * @return const val_t
   */
  inline static const val_t max_val() throw()
  { return max_quantum; }
 
 
  /**
   * Gray scale colour, where values are between 0.0 and 1.0.
   * @param float v
   * @return basic_color
   */
  inline static basic_color grayscale(float_t v) throw()
  {
    v *= max_quantum;
    v = v < 0 ? 0 : (v > max_quantum ? max_quantum : v);
    return basic_color(v,v,v);
  }
 
  /**
   * RGB color from r,g,b, where values are between 0.0 and 1.0.
   * @param float r
   * @param float g
   * @param float b
   * @param float a=0
   * @return basic_color
   */
  inline static basic_color rgb(float_t r, float_t g, float_t b, float_t a=0) throw()
  {
    return basic_color(
      max_quantum * (r>1 ? 1 : (r<=0 ? 0 : r)), 255 * (g>1 ? 1 : (g<=0 ? 0 : g)),
      max_quantum * (b>1 ? 1 : (b<=0 ? 0 : b)), 255 * (a>1 ? 1 : (a<=0 ? 0 : a))
    );
  }
 
  /**
   * RGB color from HSV input components, where values are between 0.0 and 1.0.
   * @param float h
   * @param float s
   * @param float v
   * @return basic_color
   */
  inline static basic_color hsv(float_t h, float_t s, float_t v)
  {
    s = s<0? 0 : (s>1? 1:s);
    v = v<0? 0 : (v>1? 1:v);
    h = h-((int)h)+1; h -= ((int)h); h *= 6.0;
    int a = (int)h;
    float f = h - a;
    float p = v * (1. - s);
    float q = v * (1. - s * f);
    float t = v * (1. - s * (1.-f));
    switch(a) {
      case 0: return rgb(v,t,p);
      case 1: return rgb(q,v,p);
      case 2: return rgb(p,v,t);
      case 3: return rgb(p,q,v);
      case 4: return rgb(t,p,v);
      case 5: return rgb(v,p,q);
      default:return rgb(v,t,p);
    }
  }
 
  /**
   * RGB color from HSL input components, where values are between 0.0 and 1.0.
   * @param float h
   * @param float s
   * @param float l
   * @return basic_color
   */
  inline static basic_color hsl(float_t h, float_t s, float_t l)
  { s*=(l<0.5)?(l):(1-l); return hsv(h, 2.0*s/(l+s), l+s); }
 
  /**
   * RGB heat map color from "intensity" value between 0.0 and 1.0
   * @parem float v
   * @return basic_color
   */
  inline static basic_color heatmap(float_t v)
  { v = (v<0? 0:(v>1? 1:v)); return hsv(1-(0.2+(v*0.8)), 1, 0.25+(v*0.75)); }
 
public:
 
  /**
   * Color constants
   */
  static const basic_color white; /// #000000 alpha=0
  static const basic_color black; /// #ffffff alpha=0
  static const basic_color red;   /// #ff0000 alpha=0
  static const basic_color green; /// #00ff00 alpha=0
  static const basic_color blue;  /// #0000ff alpha=0
 
protected:
 
  val_t c_[4];
};
 
template <typename val_t, typename float_t, val_t max_quantum>
std::ostream& operator<<(std::ostream& o, const basic_color<val_t,float_t,max_quantum>& c)
{ o << (std::string)c; return o; }
 
#define defcolor(NAME, VAL) \
  template <typename val_t, typename float_t, val_t m> \
  const basic_color<val_t,float_t,m> basic_color<val_t,float_t,m>::NAME(VAL)
  defcolor(black , "000");
  defcolor(white , "fff");
  defcolor(red   , "f00");
  defcolor(green , "0f0");
  defcolor(blue  , "00f");
#undef defcolor
// </editor-fold>
 
// <editor-fold desc="basic_point2d" defaultstate="collapsed">
/**
 * Point class
 */
template <typename crd_t, typename float_t>
class basic_point2d
{
public:
  inline basic_point2d() throw() : x_(0), y_(0)
  {}
 
  inline basic_point2d(crd_t x_, crd_t y_) throw() : x_(x_), y_(y_)
  {}
 
  inline basic_point2d(const basic_point2d &p) throw() : x_(p.x_), y_(p.y_)
  {}
 
  inline basic_point2d(const cv::Point2i &p) throw() : x_(p.x), y_(p.y)
  {}
 
  virtual ~basic_point2d()
  { ; }
 
  inline crd_t x() const throw()
  { return x_; }
 
  inline crd_t y() const throw()
  { return y_; }
 
  inline basic_point2d & x(crd_t x__) throw()
  { x_=x__; return *this; }
 
  inline basic_point2d & y(crd_t y__) throw()
  { y_=y__; return *this; }
 
  inline basic_point2d& operator=(const basic_point2d& p) throw()
  { x_=p.x_; y_=p.y_; return *this; }
 
  inline bool operator==(const basic_point2d& p) const throw()
  { return x_==p.x_ && y_==p.y_; }
 
  inline bool operator>=(const basic_point2d& p) const throw()
  { return x_>=p.x_ && y_>=p.y_; }
 
  inline bool operator<=(const basic_point2d& p) const throw()
  { return x_<=p.x_ && y_<=p.y_; }
 
  inline bool operator>(const basic_point2d& p) const throw()
  { return x_>p.x_ && y_>p.y_; }
 
  inline bool operator<(const basic_point2d& p) const throw()
  { return x_<p.x_ && y_<p.y_; }
 
  inline operator cv::Point2i () const
  { return cv::Point2i(x_, y_); }
 
  inline operator cv::Point2f () const
  { return cv::Point2f(x_, y_); }
 
  inline basic_point2d operator+(const basic_point2d& p) const throw()
  { return basic_point2d(x_+p.x_, y_+p.y_); }
 
  inline basic_point2d operator-(const basic_point2d& p) const throw()
  { return basic_point2d(x_-p.x_, y_-p.y_); }
 
  inline basic_point2d& operator+=(const basic_point2d& p) throw()
  { x_+=p.x_; y_+=p.y_; return *this; }
 
  inline basic_point2d& operator-=(const basic_point2d& p) throw()
  { x_-=p.x(); y_-=p.y(); return *this; }
 
  inline basic_point2d& operator*=(double d) throw()
  { x_*=d; y_*=d; return *this; }
 
  inline basic_point2d& operator/=(double d)
  { x_/=d; y_/=d; return *this; }
 
  inline double norm() const throw()
  { return sqrt((double)x_*(double)x_ + (double)y_*(double)y_); }
 
protected:
 
  crd_t x_, y_;
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_point2d<crd_t,float_t> p)
{ os << "{ x:" << p.x() << ", y:" << p.y() << " }"; return os; }
 
// </editor-fold>
 
// <editor-fold desc="basic_circle2d" defaultstate="collapsed">
/**
 * Circle class
 */
template <typename crd_t, typename float_t>
class basic_circle2d
{
public:
  typedef basic_point2d<crd_t,float_t> point_t;
 
  inline basic_circle2d() throw() : x_(0), y_(0), r_(0)
  {}
 
  inline basic_circle2d(crd_t x, crd_t y, crd_t r) throw() : x_(x),y_(y),r_(r)
  {}
 
  inline basic_circle2d(point_t p, crd_t r) throw() : x_(p.x()),y_(p.y()),r_(r)
  {}
 
  inline basic_circle2d(const basic_circle2d &c) throw() : x_(c.x_),y_(c.y_),r_(c.r_)
  {}
 
  virtual ~basic_circle2d()
  { ; }
 
  inline basic_circle2d& operator=(const basic_circle2d& c) throw()
  { x_=c.x_; y_=c.y_; r_=c.r_; return *this; }
 
  inline crd_t x() const throw()
  { return x_; }
 
  inline crd_t y() const throw()
  { return y_; }
 
  inline crd_t r() const throw()
  { return r_; }
 
  inline basic_circle2d & x(crd_t x) throw()
  { x_=x; return *this; }
 
  inline basic_circle2d & y(crd_t y) throw()
  { y_=y; return *this; }
 
  inline basic_circle2d & r(crd_t r) throw()
  { r_=r; return *this; }
 
  inline operator cv::Point () const
  { return cv::Point(x_, y_); }
 
  inline double area() const throw()
  { return r_ * r_ * M_PI; }
 
  inline basic_point2d<crd_t, float_t> center() const
  { return basic_point2d<crd_t, float_t>(x_,y_); }
 
  inline double circumference() const throw()
  { return  r_ * M_PI * 2.0; }
 
protected:
  crd_t x_, y_, r_;
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_circle2d<crd_t,float_t> c)
{ os << "{ x:" << c.x() << ", y:" << c.y() << ", r:" << c.r() << " }"; return os; }
 
// </editor-fold>
 
// <editor-fold desc="basic_line2d" defaultstate="collapsed">
/**
 * Line class
 */
template <typename crd_t, typename float_t>
class basic_line2d
{
public:
 
  inline basic_line2d() throw() : p0_(), p1_()
  {}
 
  inline basic_line2d(crd_t x0, crd_t y0, crd_t x1, crd_t y1) throw()
    : p0_(x0,y0), p1_(x1,y1)
  {}
 
  inline basic_line2d(const basic_line2d &l) throw() : p0_(l.p0_), p1_(l.p1_)
  {}
 
  inline basic_line2d(const basic_point2d<crd_t, float_t> &p0_,
    const basic_point2d<crd_t, float_t> &p1_) throw() : p0_(p0_), p1_(p1_)
  {}
 
  virtual ~basic_line2d()
  { ; }
 
  inline basic_point2d<crd_t, float_t> p0() const throw()
  { return p0_; }
 
  inline basic_point2d<crd_t, float_t> p1() const throw()
  { return p1_; }
 
  inline basic_line2d & p0(basic_point2d<crd_t, float_t> p) throw()
  { p0_=p; return *this; }
 
  inline basic_line2d & p1(basic_point2d<crd_t, float_t> p) throw()
  { p1_=p; return *this; }
 
  inline basic_line2d& operator=(const basic_line2d& l) throw()
  { p0_=l.p0_; p1_=l.p1_; return *this; }
 
  inline bool operator==(const basic_line2d& l) const throw()
  { return p0_==l.p0_ && p1_==l.p1_; }
 
  inline double length() const throw()
  { return (p1_-p0_).norm(); }
 
protected:
  basic_point2d<crd_t, float_t> p0_, p1_;
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_line2d<crd_t,float_t> l)
{ os << "{ x0:" << l.p0().x() << ", y0:" << l.p0().y()
     << ", x1:" << l.p1().x() << ", y1:" << l.p1().y()
     << " }"; return os; }
 
// </editor-fold>
 
// <editor-fold desc="basic_rect2d" defaultstate="collapsed">
/**
 * Rectangle class
 */
template <typename crd_t, typename float_t>
class basic_rect2d
{
  typedef basic_point2d<crd_t, float_t> point_t;
 
public:
 
  /**
   * Empty rect
   */
  inline basic_rect2d() throw() : p0_(), sz_()
  { ; }
 
  /**
   * 2 point component rect
   */
  inline basic_rect2d(crd_t x0, crd_t y0, crd_t x1, crd_t y1) throw()
  { p0_.x(std::min(x0,x1)); p0_.y(std::min(y0,y1)); sz_.x(std::abs(x1-x0)+1);
    sz_.y(std::abs(y1-y0)+1);
  }
 
  /**
   * Copy rect
   */
  inline basic_rect2d(const basic_rect2d &r) throw() : p0_(r.p0_), sz_(r.sz_)
  { ; }
 
  /**
   * 2 point rect, top-left, bottom-right
   */
  inline basic_rect2d(const point_t &p0, const point_t &p1) throw()
  { p0_.x(std::min(p0.x(),p1.x())); p0_.y(std::min(p0.y(),p1.y()));
    sz_.x(std::abs(p1.x()-p0.x())); sz_.y(std::abs(p1.y()-p0.y()));
  }
 
  /**
   * Point (top-left) and size components.
   */
  inline basic_rect2d(const point_t &p, crd_t width, crd_t height) throw() : p0_(p)
  { sz_.x(width<0?0:width); sz_.y(height<0?0:height); }
 
  /**
   * Copy cv::Rect
   */
  inline basic_rect2d(const cv::Rect r) throw()
  { p0_.x(r.x); p0_.y(r.y); sz_.x(r.width); sz_.y(r.height); }
 
  virtual ~basic_rect2d()
  { ; }
 
  inline point_t p() const throw()
  { return p0_; }
 
  inline point_t size() const throw()
  { return sz_; }
 
  inline crd_t width() const throw()
  { return sz_.x(); }
 
  inline void width(crd_t v) throw()
  { sz_.x(v<0?0:v); }
 
  inline crd_t height() const throw()
  { return sz_.y(); }
 
  inline void height(crd_t v) throw()
  { sz_.y(v<0?0:v); }
 
  inline crd_t x() const throw()
  { return p0_.x(); }
 
  inline void x(crd_t v) throw()
  { p0_.x(v); }
 
  inline crd_t y() const throw()
  { return p0_.y(); }
 
  inline void y(crd_t v) throw()
  { p0_.y(v); }
 
  inline point_t tl() const throw()
  { return p0_; }
 
  inline point_t br() const throw()
  { return basic_point2d<crd_t, float_t>(p0_.x()+(sz_.x()>0 ? sz_.x()-1 : 0),
              p0_.y()+(sz_.y()>0 ? sz_.y()-1 : 0)); }
 
  inline point_t center() const throw()
  { return basic_point2d<crd_t, float_t>(p0_.x()+sz_.x()/2, p0_.y()+sz_.y()/2); }
 
  inline basic_rect2d & p(point_t p) throw()
  { p0_=p; return *this; }
 
  inline basic_rect2d & size(point_t p) throw()
  { sz_=p; return *this; }
 
  inline basic_rect2d& operator=(const basic_rect2d& r) throw()
  { p0_=r.p0_; sz_=r.sz_; return *this; }
 
  inline bool operator==(const basic_rect2d& r) const throw()
  { return p0_ == r.p0_ && sz_ == r.sz_; }
 
  inline operator cv::Rect () const
  { return cv::Rect((cv::Point2i) p0_, (cv::Size2i(sz_.x(), sz_.y()))); }
 
  inline double area() const throw()
  { return sz_.x()*sz_.y(); }
 
  inline double circumference() const throw()
  { return  sz_.x() * 2.0 + sz_.y() * 2.0; }
 
  inline operator std::vector<cv::Point2i> () const
  {
    std::vector<cv::Point2i> v;
    v.push_back(cv::Point2i(p0_.x(), p0_.y()));
    v.push_back(cv::Point2i(p0_.x()+sz_.x(), p0_.y()));
    v.push_back(cv::Point2i(p0_.x()+sz_.x(), p0_.y()+sz_.y()));
    v.push_back(cv::Point2i(p0_.x(), p0_.y()+sz_.y()));
    v.push_back(cv::Point2i(p0_.x(), p0_.y()));
    return v;
  }
 
protected:
 
  basic_point2d<crd_t, float_t> p0_, sz_;
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_rect2d<crd_t,float_t> l)
{ os << "{ x:" << l.x() << ", y:" << l.y() << ", w:" << l.width() << ", h:"
    << l.height() << " }"; return os; }
 
// </editor-fold>
 
// <editor-fold desc="basic_rotated_rect2d" defaultstate="collapsed">
/**
 * Rotates rectangle class
 */
template <typename crd_t, typename float_t>
class basic_rotated_rect2d
{
public:
 
  inline basic_rotated_rect2d() throw() : c_(), sz_(), a_(0)
  {}
 
  inline basic_rotated_rect2d(crd_t x, crd_t y, crd_t w, crd_t h, float_t a=0)
    throw() : c_(x,y), sz_(w,h), a_(a)
  {}
 
  inline basic_rotated_rect2d(const basic_rotated_rect2d &r) throw()
    : c_(r.c_), sz_(r.sz_), a_(r.a_)
  {}
 
  inline basic_rotated_rect2d(const basic_point2d<crd_t, float_t> &c,
    const basic_point2d<crd_t, float_t> &sz, float_t a=0) throw() : c_(c), sz_(sz),
    a_(a)
  {}
 
  inline basic_rotated_rect2d(const cv::RotatedRect &r) throw() : c_(r.center)
  { sz_.x(r.size.width); sz_.y(r.size.height); a_ = r.angle; }
 
  virtual ~basic_rotated_rect2d()
  { ; }
 
  inline basic_point2d<crd_t, float_t> center() const throw()
  { return c_; }
 
  inline basic_point2d<crd_t, float_t> size() const throw()
  { return sz_; }
 
  inline crd_t angle() const throw()
  { return a_; }
 
  inline basic_rotated_rect2d & center(const basic_point2d<crd_t, float_t>& p) throw()
  { c_ = p; return *this; }
 
  inline basic_rotated_rect2d & size(const basic_point2d<crd_t, float_t>& sz) throw()
  { sz_ = sz; return *this; }
 
  inline basic_rotated_rect2d & angle(crd_t a) const throw()
  { a_=a; return *this; }
 
  inline basic_rotated_rect2d& operator=(const basic_rotated_rect2d& r) throw()
  { c_=r.c_; sz_=r.sz_; a_=r.a_; return *this; }
 
  inline operator cv::Point () const throw()
  { return c_; }
 
  inline operator cv::RotatedRect () const
  { return cv::RotatedRect(cv::Point2f(c_.x(), c_.y()), cv::Size2f(sz_.x(), sz_.y()), a_); }
 
  inline double area() const throw()
  { return sz_.x()*sz_.y(); }
 
  inline double circumference() const throw()
  { return  sz_.x() * 2.0 + sz_.y() * 2.0; }
 
  inline operator std::vector<cv::Point2i> () const
  {
    std::vector<cv::Point2i> v;
    cv::Point2f rpt[4];
    ((cv::RotatedRect)*this).points(rpt);
    for(int i=0; i<4; ++i) v.push_back(cv::Point2i(rpt[i].x,rpt[i].y));
    return v;
  }
 
protected:
 
  basic_point2d<crd_t, float_t> c_, sz_;
  float_t a_;
};
 
// </editor-fold>
 
// <editor-fold desc="basic_ellipse2d" defaultstate="collapsed">
/**
 * Ellipse class
 */
template <typename crd_t, typename float_t>
class basic_ellipse2d
{
public:
 
  inline basic_ellipse2d() throw() : c_(), sz_(), a_(0)
  {}
 
  inline basic_ellipse2d(crd_t x, crd_t y, crd_t w, crd_t h, float_t a=0) throw()
    : c_(x,y), sz_(w,h), a_(a)
  {}
 
  inline basic_ellipse2d(const basic_ellipse2d &r) throw() : c_(r.c_), sz_(r.sz_),
    a_(r.a_)
  {}
 
  inline basic_ellipse2d(const basic_point2d<crd_t, float_t> &c, const
    basic_point2d<crd_t, float_t> &sz, float_t a=0) throw() : c_(c), sz_(sz), a_(a)
  {}
 
  virtual ~basic_ellipse2d()
  { ; }
 
  inline basic_point2d<crd_t, float_t> center() const throw()
  { return c_; }
 
  inline basic_point2d<crd_t, float_t> size() const throw()
  { return sz_; }
 
  inline crd_t angle() const throw()
  { return a_; }
 
  inline basic_ellipse2d & center(const basic_point2d<crd_t, float_t>& p) throw()
  { c_ = p; return *this; }
 
  inline basic_ellipse2d & size(const basic_point2d<crd_t, float_t>& sz) throw()
  { sz_ = sz; return *this; }
 
  inline basic_ellipse2d & angle(crd_t a) const throw()
  { a_=a; return *this; }
 
  inline basic_ellipse2d& operator=(const basic_ellipse2d& r) throw()
  { c_=r.c_; sz_=r.sz_; a_=r.a_; return *this; }
 
  inline operator cv::Point () const
  { return c_; }
 
  inline operator cv::RotatedRect () const
  { return cv::RotatedRect(cv::Point2f(c_.x(), c_.y()), cv::Size2f(sz_.x(), sz_.y()), a_); }
 
  inline double area() const throw()
  { return  sz_.x() * sz_.y() * M_PI/4; }
 
  inline double circumference() const throw()
  { return (double)2.0 * M_PI * sqrt((sz_.x()/2*sz_.x()/2)+(sz_.y()/2)*(sz_.y()/2)); }
 
protected:
 
  basic_point2d<crd_t, float_t> c_, sz_;
  float_t a_;
};
// </editor-fold>
 
// <editor-fold desc="basic_polynom2d" defaultstate="collapsed">
/**
 * Polynom
 */
template <typename crd_t, typename float_t>
class basic_polynom2d : public std::vector<basic_point2d<crd_t,float_t> >
{
public:
 
  typedef basic_point2d<crd_t,float_t> point_t;
  typedef basic_rect2d<crd_t,float_t> rect_t;
  typedef typename std::vector<point_t> vect_t;
 
  inline basic_polynom2d() : vect_t()
  { ; }
 
  inline basic_polynom2d(const point_t &p) : vect_t()
  { vect_t::push_back(p); }
 
  inline basic_polynom2d(const rect_t &r) : vect_t()
  {
    vect_t::push_back(point_t(0,0));
    vect_t::push_back(point_t(r.br().x(), 0));
    vect_t::push_back(point_t(r.br().x(), r.br().y() ));
    vect_t::push_back(point_t(0, r.br().y()));
    vect_t::push_back(point_t(0,0));
  }
 
  inline basic_polynom2d(const std::vector<cv::Point2i> &p) : vect_t()
  {
    if(p.empty()) return;
    vect_t::reserve(p.size());
    for(typename std::vector<cv::Point2i>::const_iterator it=p.begin(); it!=p.end(); ++it) {
      vect_t::push_back(point_t(*it));
    }
  }
 
  virtual ~basic_polynom2d()
  { ; }
 
  inline basic_polynom2d& add(const point_t &p)
  { vect_t::push_back(p); return *this; }
 
  inline basic_polynom2d& add(const basic_polynom2d &p)
  {
    if(p.empty()) return *this;
    typename vect_t::const_iterator it=p.begin();
    if(!vect_t::empty() && *(vect_t::rbegin()) == *(p.begin())) ++it;
    while(it!=p.end()) { vect_t::push_back(*it); ++it; }
    return *this;
  }
 
  inline basic_polynom2d& operator=(const point_t &p)
  { vect_t::clear(); push_back(p); return *this; }
 
  inline basic_polynom2d& operator+=(const point_t &p)
  { return add(p); }
 
  inline basic_polynom2d& operator+=(const basic_polynom2d &p)
  { return add(p); }
 
  /**
   * Returns true if the polynom is closed
   * @return
   */
  virtual bool closed() const
  {
    if(vect_t::empty()) return false;
    point_t p = *vect_t::begin(); p-=*vect_t::rbegin();
    return p.norm() < 1.5;
  }
 
  /**
   * To cv::Point2i vector
   * @return std::vector<cv::Point2i>
   */
  inline operator std::vector<cv::Point2i> () const
  {
    std::vector<cv::Point2i> v;
    for(typename vect_t::const_iterator it=vect_t::begin(); it!=vect_t::end(); ++it) {
      v.push_back((cv::Point2i)*it);
    }
    return v;
  }
 
  /**
   * Approximate lower resolution polynom with defined maximum error.
   * @param float_t max_error=1
   * @return basic_polynom2d&
   */
  inline basic_polynom2d& approx(float_t max_error=1)
  {
    std::vector<cv::Point2i> v = *this, v2;
    (*this).clear();
    cv::approxPolyDP(v, v2, max_error, closed());
    return ((*this) = basic_polynom2d(v2));
  }
 
  /**
   * Returns true if the polynom is convex
   * @return bool
   */
  inline bool convex() const
  { return cv::isContourConvex(cv::Mat((std::vector<cv::Point2i>)(*this))); }
 
  /**
   * Returns the perimeter / arc length of the polynom
   * @return double
   */
  inline double perimeter() const
  {
    if(vect_t::size() < 2) return 0;
    typename vect_t::const_iterator p0 = vect_t::begin();
    typename vect_t::const_iterator p1 = vect_t::begin()+1;
    double len = closed() ? 0 : (vect_t::front()-vect_t::back()).norm();
    for(; p1 != vect_t::end(); ++p0, ++p1) len += (*p1-*p0).norm();
    return len;
  }
 
  /**
   * Returns the perimeter / arc length of the polynom
   * @return double
   */
  inline double arc_length() const
  { return perimeter(); }
 
  /**
   * Returns the arc length of the polynom
   * @return
   */
  inline rect_t bounding_rect() const
  {
    if(vect_t::empty()) return rect_t();
    crd_t x1, y1, x0, y0;
    x1 = x0 = vect_t::front().x();
    y1 = y0 = vect_t::front().y();
    for(typename vect_t::const_iterator it = vect_t::begin(); it != vect_t::end(); ++it) {
      if(it->x() > x1) x1 = it->x(); else if(it->x() < x0) x0 = it->x();
      if(it->y() > y1) y1 = it->y(); else if(it->y() < y0) y0 = it->y();
    }
    return rect_t(x0, y0, x1, y1);
  }
 
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_polynom2d<crd_t,float_t> p)
{
  if(p.empty()) return os;
  os << '[' << (*p.begin());
  for(typename basic_polynom2d<crd_t,float_t>::const_iterator it=p.begin()+1; it!=p.end(); ++it) {
    os << "," << (*it);
  }
  os << ']';
  return os;
}
 
// </editor-fold>
 
// <editor-fold desc="basic_contour2d" defaultstate="collapsed">
/**
 * Contour
 */
template <typename crd_t, typename float_t>
class basic_contour2d : public basic_polynom2d<crd_t,float_t>
{
public:
  typedef basic_point2d<crd_t,float_t> point_t;
  typedef basic_polynom2d<crd_t,float_t> poly_t;
  typedef basic_rect2d<crd_t,float_t> rect_t;
  typedef typename std::vector<point_t> vect_t;
 
  inline basic_contour2d() : poly_t()
  { }
 
  inline basic_contour2d(const point_t &p) : poly_t(p)
  { }
 
  inline basic_contour2d(const rect_t &r) : basic_polynom2d<crd_t,float_t>(r)
  { }
 
  inline basic_contour2d(const std::vector<cv::Point2i> &p) : basic_polynom2d<crd_t,float_t>(p)
  { }
 
  virtual ~basic_contour2d()
  { ; }
 
  /**
   * Returns true if the contour is closed (always true for contours not for the polygon base class)
   * @return bool
   */
  virtual bool closed() const throw()
  { return true; }
 
  /**
   * Returns the area of the contour
   * @param bool oriented
   * @return crd_t
   */
  inline crd_t area(bool oriented=false) const
  { return cv::contourArea(cv::Mat((std::vector<cv::Point2i>)(*this)), oriented); }
 
};
 
template <typename crd_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_contour2d<crd_t,float_t> c)
{ return operator<< (os, (basic_polynom2d<crd_t,float_t>) c); }
 
// </editor-fold>
 
// <editor-fold desc="basic_contour_tree" defaultstate="collapsed">
/**
 * Contour tree. As the class uses pointers internally, you cannot add, insert
 * or erase, only clear node contents and children using clear().
 */
template <typename crd_t, typename val_t, typename float_t>
class basic_contour_tree
{
  friend class basic_image<crd_t, val_t, float_t>;
 
public: // Types
 
  typedef basic_contour2d<crd_t, float_t> elem_t;
  typedef std::vector<basic_contour_tree> container_t;
 
public: // Initialisers
 
  inline basic_contour_tree() : elem_(), children_(), parent_(0), level_(0)
  { ; }
 
  inline basic_contour_tree(elem_t e) : elem_(e), children_(), parent_(0), level_(0)
  { ; }
 
  virtual ~basic_contour_tree()
  { clear(); }
 
public: // Conversions, info
 
  inline bool empty() const throw()
  { return children_.empty() && elem_.empty(); }
 
  inline bool has_children() const throw()
  { return !children_.empty(); }
 
  inline bool has_parent() const throw()
  { return parent_!=0; }
 
  inline const basic_contour_tree& parent() const
  { return parent_!=0 ? *parent_ : *this; }
 
  inline const elem_t& val() const throw()
  { return elem_; }
 
  inline elem_t& val() throw()
  { return elem_; }
 
  inline operator const elem_t& () const throw()
  { return elem_; }
 
  inline const container_t& children() const throw()
  { return children_; }
 
  inline operator const container_t& () const throw()
  { return children_; }
 
  inline void clear()
  { children_.clear(); elem_.clear(); }
 
protected:
 
  elem_t elem_;
  container_t children_;
  basic_contour_tree* parent_;
  int level_;
};
// </editor-fold>
 
/**
 * Class basic_image
 */
template <typename crd_t, typename val_t, typename float_t>
class basic_image
{
public:
 
  // <editor-fold desc="types" defaultstate="collapsed">
  typedef typename cv::Mat mat_t; // dev: can be added to template args
  typedef typename std::vector<cv::Mat> mat_vector_t;
  typedef char char_t;  // dev: can be added to template args
 
  typedef enum { ls_no=8, ls_connected=4, ls_antialiased=CV_AA, ls_filled=CV_FILLED } line_style_type;
  typedef enum { csp_gray=0, csp_rgb=1, csp_hsv=2 } colorspace_t;
  typedef enum { load_as_is=0, load_gray, load_color } load_mode_t;
 
  typedef basic_image_exception<char_t> exception;
  typedef basic_color<val_t, float_t> color_t;
  typedef basic_point2d<crd_t, float_t> point_t;
  typedef basic_line2d<crd_t, float_t> line_t;
  typedef basic_rect2d<crd_t, float_t> rect_t;
  typedef basic_circle2d<crd_t, float_t> circle_t;
  typedef basic_rotated_rect2d<crd_t, float_t> rotated_rect_t;
  typedef basic_ellipse2d<crd_t, float_t> ellipse_t;
  typedef basic_polynom2d<crd_t, float_t> polynom_t;
  typedef basic_contour2d<crd_t, float_t> contour_t;
  typedef std::vector<line_t> line_list_t;
  typedef std::vector<circle_t> circle_list_t;
  typedef std::vector<contour_t> contour_list_t;
  typedef struct { contour_list_t outer, inner; } contour_ccomp_list_t;
  typedef basic_contour_tree<crd_t,val_t,float_t> contour_nested_list_t;
  typedef basic_histogram<crd_t,val_t,float_t> histogram_t;
  // </editor-fold>
 
public:
 
  // <editor-fold desc="constructors / assignment, type conversion" defaultstate="collapsed">
 
  // Constructors: Cloning ensures continuous cv::Mat's
 
  /**
   * Standard
   */
  basic_image() throw() : m_(), file_(), is_hsv_(false)
  { ; }
 
  /**
   * Copy constructor, clones the cv::Mat
   * @param const basic_image &im
   */
  basic_image(const basic_image &im)  : m_(im.m_.clone()), file_(), is_hsv_(false)
  { ; }
 
  /**
   * Copy constructor from cv::Mat, clones the cv::Mat
   * @param const mat_t &m
   */
  basic_image(const mat_t &m) : m_(m.clone()), file_(), is_hsv_(false)
  { ; }
 
  /**
   * Constructor by size and color space
   * @param crd_t width
   * @param crd_t height
   * @param colorspace_t colorsp
   */
  inline basic_image(crd_t width, crd_t height, colorspace_t colorsp) : m_(), file_(),
          is_hsv_(colorsp==csp_hsv)
  { m_ = cv::Mat::zeros((int)height, (int)width, colorsp == csp_gray ? CV_8UC1 : CV_8UC3); }
 
  /**
   * Constructor by size and color space and predefined "background color".
   * @param crd_t width
   * @param crd_t height
   * @param colorspace_t colorsp
   */
  basic_image(crd_t width, crd_t height, colorspace_t colorsp, const color_t & color) :
    m_(), file_(), is_hsv_(colorsp==csp_hsv)
  {
    m_.create((int)height, (int)width, colorsp == csp_gray ? CV_8UC1 : CV_8UC3);
    m_ = (cv::Scalar) color;
  }
 
  /**
   * Constructor by file
   * @param const std::string& file
   */
  inline basic_image(const std::string& file, load_mode_t mode=load_as_is) : m_(), file_(),
          is_hsv_(false)
  { load(file, mode); }
 
  virtual ~basic_image()
  { ; }
 
  inline basic_image& operator=(const basic_image& im)
  { m_ = im.mat().clone(); is_hsv_ = im.is_hsv_; return *this; }
 
  inline basic_image& operator=(const mat_t& m)
  { m_ = m.clone(); is_hsv_=false; return *this; }
 
  inline mat_t& mat() throw()
  { return m_; }
 
  inline const mat_t& mat() const throw()
  { return m_; }
 
  inline operator mat_t& () throw()
  { return m_; }
 
  inline operator const mat_t& () throw()
  { return m_; }
 
  inline basic_image& mat(const mat_t& m)
  { m_ = m.clone(); return *this; }
 
  // </editor-fold>
 
public: // Import / export
 
  // <editor-fold desc="load/save" defaultstate="collapsed">
 
  /**
   * Load an image
   * @param const std::string& file
   * @param load_mode mode=load_as_is
   * @return basic_image&
   */
  inline basic_image& load(const std::string& file, load_mode_t mode=load_as_is)
  {
    try {
      file_ = "";
      m_ = mat_t();
      m_ = cv::imread(file, (
        mode == load_as_is ? CV_LOAD_IMAGE_UNCHANGED : (
        mode == load_gray ? CV_LOAD_IMAGE_GRAYSCALE :
        CV_LOAD_IMAGE_COLOR
      )));
      file_ = file;
      is_hsv_ = false;
    } catch(...) {
      iex("Failed to load image.");
    }
    return *this;
  }
 
  /**
   * Returns the file the image was loaded from.
   * @param file const std::string&
   * @return
   */
  inline const std::string& file() const throw()
  { return file_; }
 
  /**
   * Save the image as RGB or gray scale, depending on the image with the file
   * name that it was loaded with.
   * @return basic_image&
   */
  inline basic_image& save()
  { save(file_); return *this; }
 
  /**
   * Save the image as RGB or gray scale, depending on the image. Does not modify the
   * loaded file path retrieved with file(), except this path is empty (image is not a loaded
   * file).
   * @param const std::string &file
   * @return basic_image&
   */
  inline basic_image& save(const std::string &file)
  {
    try{
      if(!empty()) { if(is_hsv_) to_rgb(); cv::imwrite(file, m_); }
      if(file_.empty()) file_ = file;
    } catch(...) {
      iex("Failed to save image.");
    }
    return *this;
  }
 
  // </editor-fold>
 
public: // Image information
 
  // <editor-fold desc="general getters" defaultstate="collapsed">
  /**
   * Return maximum quantum value
   * @return double
   */
  inline const val_t max_quant() const throw()
  { return color_t::max_val(); }  // CHECK more common way: _m.elemSize1() or the like
 
  /**
   * Returns if the image is empty.
   * @return
   */
  inline bool empty() const throw()
  { return !m_.data || m_.cols == 0 || m_.rows == 0; }
 
  /**
   * Returns the number of image channels.
   * @return unsigned
   */
  inline unsigned channels() const throw()
  { return m_.channels(); }
 
  /**
   * Image width
   * @return crd_t
   */
  inline crd_t width() const throw()
  { return m_.cols; }
 
  /**
   * Image height
   * @return crd_t
   */
  inline crd_t height() const throw()
  { return m_.rows; }
 
  /**
   * Image area
   * @return crd_t
   */
  inline crd_t area() const throw()
  { return m_.rows * m_.cols; }
 
  /**
   * Image size rectangle
   * @return rect_t
   */
  inline point_t size() const throw()
  { return point_t(width(), height()); }
 
  /**
   * Returns if the image color space is currently RGB (t.m. matrix is BGR, 3ch)
   * @return bool
   */
  inline bool is_rgb() const throw()
  { return !is_hsv_ && channels()>1; }
 
  /**
   * Returns if the image color space is currently gray scale (1 channel).
   * @return bool
   */
  inline bool is_gray() const throw()
  { return channels()<2; }
 
  /**
   * Returns if the image color space is currently HSV (3 channels)
   * @return bool
   */
  inline bool is_hsv() const throw()
  { return is_hsv_ && channels()>1; }
 
  /**
   * Returns the current color space of the image
   * @return colorspace_t
   */
  inline colorspace_t color_space() const throw()
  { return (channels()<2) ? csp_gray : (is_hsv_ ? csp_hsv : csp_rgb); }
 
  /**
   * Sets the current color space of the image
   * @param colorspace_t
   */
  inline basic_image& color_space(colorspace_t c)
  {
    if(color_space() == c) return *this;
    else if(c == csp_gray) return to_grayscale();
    else if(c == csp_rgb) return to_rgb();
    else if(c == csp_hsv) return to_hsv();
    return *this;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="num_nonzero_pixels, nonzero_pixels" defaultstate="collapsed">
  /**
   * Returns the number of nonzero pixels. If the image is not gray scale,
   * a temporary gray scale copy is created (performance!).
   * @return long
   */
  inline long num_nonzero_pixels() const
  {
    return (channels()<2) ? cv::countNonZero(m_) :
      basic_image(*this).to_grayscale().num_nonzero_pixels();
  }
 
 /**
   * Return nonzero pixel coordinates.
   * Note: E.g.
   *
   *    struct my_point_type; // MUST have constructor my_point_type(x, y)
   *    std::list<my_point_type> pts;
   *    unsigned n = img.unsigned nonzero_pixels(pts);
   *
   * @param typename point_list_type& points
   * @return unsigned
   */
  template <typename point_list_type>
  inline unsigned nonzero_pixels(point_list_type& points) const
  {
    points.clear();
    basic_image im = *this;
    im.to_grayscale();
    unsigned np = im.num_nonzero_pixels();
    if(!np) return 0;
    register unsigned n = np;
    for(register int iy=im.height()-1; iy >= 0; --iy)  {
      for(register int ix=im.width()-1; ix >= 0; --ix)  {
        if(im.gray_value_at(ix, iy)) {
          points.push_back(typename point_list_type::value_type(ix,iy));
          if(--n <= 0) return np;
        }
      }
    }
    return np;
  }
 
 /**
   * Return nonzero pixel coordinates. Specialised version with `reserve()` call.
   * Note: E.g.
   *
   *    struct my_point_type; // MUST have constructor my_point_type(x, y)
   *    std::vector<my_point_type> pts;
   *    unsigned n = img.unsigned nonzero_pixels(pts);
   *
   * @param std::vector<point_type>& points
   * @return unsigned
   */
  template <typename point_type>
  inline unsigned nonzero_pixels(std::vector<point_type>& points) const
  {
    points.clear();
    basic_image im = *this;
    im.to_grayscale();
    unsigned np = im.num_nonzero_pixels();
    if(!np) return 0;
    points.reserve(np);
    register unsigned n = np;
    for(register int iy=im.height()-1; iy >= 0; --iy)  {
      for(register int ix=im.width()-1; ix >= 0; --ix)  {
        if(im.gray_value_at(ix, iy)) {
          points.push_back(point_type(ix,iy));
          if(--n <= 0) return np;
        }
      }
    }
    return np;
  }
  // </editor-fold>
 
  // <editor-fold desc="bounding_rect" defaultstate="collapsed">
  /**
   * Determines the bounding rectangle of the image. T.m. the image content
   * rectangle without the border. Uses x/y Sobel gradient.
   * @param float_t gradient_threshold=0.2
   * @param int gradient_radius=0
   * @return rect_t
   */
  inline rect_t bounding_rect(float_t gradient_threshold, int gradient_radius=0) const
  {
    // +++ check if finding external contours is faster +++
    basic_image im(*this);
    im.gradient(gradient_radius).threshold(gradient_threshold);
    int x0=0, x1=im.width()-1, y0=0, y1=im.height()-1;
    while(y1>0 && cv::sum(im.mat().row(y1))[0]==0) y1--;
    while(y0<y1 && cv::sum(im.mat().row(y0))[0]==0) y0++;
    while(x1>0 && cv::sum(im.mat().col(x1))[0]==0) x1--;
    while(x0<x1 && cv::sum(im.mat().col(x0))[0]==0) x0++;
    return rect_t(x0, y0, x1, y1);
  }
 
  /**
   * Determines the bounding rectangle of the image. All nonzero pixels count.
   * @return rect_t
   */
  inline rect_t bounding_rect() const
  {
    basic_image im(*this);
    im.threshold(0.1);
    int x0=0, x1=im.width()-1, y0=0, y1=im.height()-1;
    while(y1>0 && cv::sum(im.mat().row(y1))[0]==0) y1--;
    while(y0<y1 && cv::sum(im.mat().row(y0))[0]==0) y0++;
    while(x1>0 && cv::sum(im.mat().col(x1))[0]==0) x1--;
    while(x0<x1 && cv::sum(im.mat().col(x0))[0]==0) x0++;
    return rect_t(x0, y0, x1, y1);
  }
  // </editor-fold>
 
  // <editor-fold desc="min, max, mean" defaultstate="collapsed">
  /**
   * Returns the minimum value in the image matrix (each channel separately).
   * @return val_t
   */
  inline color_t min() const
  {
    if(is_gray()) {
      double d; cv::minMaxLoc(m_, &d, 0); return color_t((int)d);
    } else {
      color_t c; double d;
      mat_vector_t chs = split_channels();
      for(unsigned i=0; i<3&&i<chs.size(); ++i) {
        cv::minMaxLoc(chs[i], &d, 0); c[i] = d;
      }
      return c;
    }
  }
 
  /**
   * Returns the minimum value in the image matrix (each channel separately).
   * @return val_t
   */
  inline color_t max() const
  {
    if(is_gray()) {
      double d; cv::minMaxLoc(m_, 0, &d); return color_t((int)d);
    } else {
      color_t c; double d;
      mat_vector_t chs = split_channels();
      for(unsigned i=0; i<3&&i<chs.size(); ++i) {
        cv::minMaxLoc(chs[i], 0, &d); c[i] = d;
      }
      return c;
    }
  }
 
  /**
   * Returns the mean "colour" of all image pixels. If the image is gray scale,
   * the all R,G,B have the same value.
   * @return color_t
   */
  inline color_t mean() const
  { return is_gray() ? color_t(cv::mean(m_)[0]) : color_t(cv::mean(m_)); }
 
  /**
   * Returns the mean "colour" of all image pixels. If the image is gray scale,
   * the all R,G,B have the same value.
   * @return color_t
   */
  inline color_t mean(const basic_image& mask) const
  {
    cv::Mat msk;
    if(!mask.empty()) msk = mask.mat();
    return is_gray() ? color_t(cv::mean(m_, msk)[0]) : color_t(cv::mean(m_, msk));
  }
  // </editor-fold>
 
  // <editor-fold desc="color_at, color_mean" defaultstate="collapsed">
  /**
   * Returns colour at given coordinates.
   * NOTE: This method is NOT suitable for large scale iterations, as it needs
   * to determine the matrix channels/depth every time it is called (switch()).
   * @param crd_t x
   * @param crd_t y
   * @return color_t
   */
  color_t color_at(crd_t x, crd_t y) const
  {
    color_t c;
    if(channels()!=1 && channels() != 3) {
      iex("color_at(): Unsupported number of channels (must be 1 or 3).");
    } else if(x<0 || y<0 || x>=width() || y>=height()) {
      return color_t();
    }
    switch(m_.depth()) {
      case CV_8U:
        if(channels()==1) {
          c = color_t(m_.at<unsigned char>(y,x));
        } else {
          cv::Vec3b v(m_.at<cv::Vec3b>(y,x));
          c = color_t(v[2],v[1],v[0]);
        }
        break;
      case CV_32F:
        if(channels()==1) {
          c = color_t((unsigned)m_.at<float>(y,x));
        } else {
          cv::Vec3f v(m_.at<cv::Vec3f>(y,x));
          c = color_t(v[2],v[1],v[0]);
        }
        break;
      case CV_64F:
        if(channels()==1) {
          c = color_t((unsigned)m_.at<double>(y,x));
        } else {
          cv::Vec3d v(m_.at<cv::Vec3d>(y,x));
          c = color_t(v[2],v[1],v[0]);
        }
        break;
      default:
        iex("color_at(): Unsupported depth.");
    }
    return c;
  }
 
 /**
   * Sets the color at a given pixel.
   * NOTE: This method is NOT suitable for large scale iterations, as it needs
   * to determine the matrix channels/depth every time it is called (switch()).
   * @param crd_t x
   * @param crd_t y
   * @param const color_t& c
   * @return color_t
   */
  basic_image& color_at(crd_t x, crd_t y, const color_t& c)
  {
    if(channels()!=1 && channels() != 3) {
      iex("color_at(): Unsupported number of channels (must be 1 or 3).");
    } else if(x<0 || y<0 || x>=width() || y>=height()) {
      return *this;
    }
    switch(m_.depth()) {
      case CV_8U:
        if(channels()==1) {
          m_.at<unsigned char>(y,x) = (((int)c[0]) + c[1] + c[2])/3;
        } else {
          m_.at<cv::Vec3b>(y,x) = cv::Vec3b(c[2], c[1], c[0]);
        }
        break;
      case CV_32F:
        if(channels()==1) {
          m_.at<float>(y,x) = (((float)c[0]) + c[1] + c[2])/3;
        } else {
          m_.at<cv::Vec3f>(y,x) = cv::Vec3f(c[2], c[1], c[0]);
        }
        break;
      case CV_64F:
        if(channels()==1) {
          m_.at<double>(y,x) = (((double)c[0]) + c[1] + c[2])/3;
        } else {
          m_.at<cv::Vec3d>(y,x) = cv::Vec3d(c[2], c[1], c[0]);
        }
        break;
      default:
        iex("color_at(): Unsupported depth.");
    }
    return *this;
  }
 
 /**
   * Sets the gray scale value at a given pixel.
   * NOTE: This method is NOT suitable for large scale iterations, as it needs
   * to determine the matrix channels/depth every time it is called (switch()).
   * @param crd_t x
   * @param crd_t y
   * @param val_t c
   * @return color_t
   */
  basic_image& color_at(crd_t x, crd_t y, val_t c)
  {
    if(channels()!=1 && channels() != 3) {
      iex("color_at(): Unsupported number of channels (must be 1 or 3).");
    } else if(x<0 || y<0 || x>=width() || y>=height()) {
      return *this;
    }
    switch(m_.depth()) {
      case CV_8U:
        if(channels()==1) {
          m_.at<unsigned char>(y,x) = c;
        } else {
          m_.at<cv::Vec3b>(y,x) = cv::Vec3b(c,c,c);
        }
        break;
      case CV_32F:
        if(channels()==1) {
          m_.at<float>(y,x) = c;
        } else {
          m_.at<cv::Vec3f>(y,x) = cv::Vec3f(c,c,c);
        }
        break;
      case CV_64F:
        if(channels()==1) {
          m_.at<double>(y,x) = c;
        } else {
          m_.at<cv::Vec3d>(y,x) = cv::Vec3d(c,c,c);
        }
        break;
      default:
        iex("color_at(): Unsupported depth.");
    }
    return *this;
  }
 
  /**
   * Returns colour at given point coordinates
   * @param const point_t &p
   * @return color_t
   */
  inline color_t color_at(const point_t &p) const
  { return color_at(p.x(), p.y()); }
 
 /**
   * Returns the gray value at a given pixel or -1 if the image is not gray.
   * @param crd_t x
   * @param crd_t y
   * @return int
   */
  inline int gray_value_at(crd_t x, crd_t y) const
  {
    #if defined(DEBUG) || defined(_DEBUG) || defined(_DEBUG_)
    return (channels() !=1 || m_.depth() != CV_8U) ? -1 : m_.at<unsigned char>(y,x);
    #else
    return m_.at<unsigned char>(y,x);
    #endif
  }
 
  // </editor-fold>
 
public: // Image (matrix) conversions, channel selection
 
  // <editor-fold desc="color conversion" defaultstate="collapsed">
  /**
   * Converts the image to gray scale (if not yet gray scale)
   * @return basic_image&
   */
  inline basic_image& to_grayscale()
  {
    if(m_.type() != CV_8U) m_.convertTo(m_, CV_8U);
    if(channels() > 1) {
      if(is_hsv_) cv::cvtColor(m_, m_, CV_HSV2BGR);
      cv::cvtColor(m_, m_, CV_BGR2GRAY);
    }
    is_hsv_ = false;
    return *this;
  }
 
  /**
   * Converts the image to BGR, 8 bits per channel (if not yet gray scale)
   * @return basic_image&
   */
  inline basic_image& to_rgb()
  {
    if(m_.type() != CV_8U) m_.convertTo(m_, CV_8U);
    if(channels() > 1 && is_hsv_) cv::cvtColor(m_, m_, CV_HSV2BGR);
    if(channels() == 1) cv::cvtColor(m_, m_, CV_GRAY2BGR);
    is_hsv_ = false;
    return *this;
  }
 
  /**
   * Converts image to HSV, 8 bits per channel
   * @return basic_image&
   */
  inline basic_image& to_hsv()
  {
    if(m_.type() != CV_8U) m_.convertTo(m_, CV_8U);
    if(channels() == 1) cv::cvtColor(m_, m_, CV_GRAY2BGR);
    if(is_hsv_) return *this;
    cv::cvtColor(m_, m_, CV_BGR2HSV);
    is_hsv_ = true;
    return *this;
  }
  // </editor-fold>
 
  // <editor-fold desc="channel extraction / composition" defaultstate="collapsed">
  /**
   * Splits image into the three RGB channels, each one gray scale image.
   * @return mat_vector;
   */
  inline mat_vector_t split_channels() const
  { mat_vector_t v; cv::split(m_, v); return v; }
 
  /**
   * Merges channels in to this image.
   * @param const mat_vector_t& channels;
   * @return basic_image& (*this)
   */
  inline basic_image& merge_channels(const mat_vector_t& chs)
  { cv::merge(chs, m_); return *this; }
 
  /**
   * Splits image into the three HSV channels, each one gray scale image.
   * @return mat_vector;
   */
  inline mat_vector_t split_channels_hsv() const
  {
    basic_image im = *this;
    im.to_rgb();
    cv::cvtColor(im.mat(), im.mat(), CV_BGR2HSV);
    mat_vector_t v;
    cv::split(im.mat(), v);
    return v;
  }
 
  /**
   * Returns RGB B channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_blue() const
  { return (channels()==1) ? basic_image(m_) : basic_image(split_channels()[0]); }
 
  /**
   * Returns RGB G channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_green() const
  { return (channels()==1) ? basic_image(m_) : basic_image(split_channels()[1]); }
 
  /**
   * Returns RGB R channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_red() const
  { return (channels()==1) ? basic_image(m_) : basic_image(split_channels()[2]); }
 
  /**
   * Returns BGR or HSV H channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_hsv_hue() const
  { return is_hsv_ ? split_channels()[0] : split_channels_hsv()[0]; }
 
  /**
   * Returns BGR or HSV S channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_hsv_sat() const
  { return is_hsv_ ?  split_channels()[1] : split_channels_hsv()[1]; }
 
  /**
   * Returns BGR or HSV V channel as gray scale image
   * @return basic_image
   */
  inline basic_image channel_hsv_val() const
  { return is_hsv_ ?  split_channels()[2] : split_channels_hsv()[2]; }
 
  // </editor-fold>
 
public: // Methods returning a copy / new image
 
  // <editor-fold desc="subimage, in_color_range" defaultstate="collapsed">
  /**
   * Returns the sub image (a COPY) specified by the rectangle `r`.
   * @param const rect_t &r
   * @return basic_image
   */
  inline basic_image subimage(const rect_t &r) const
  { return m_((cv::Rect)r).clone(); }
 
  /**
   * Creates a mask where all pixels that are in range of the specified colour
   * are nonzero, all other colours are black (0). The mask is a CLONED image.
   * The difference is normed to 0..1 (for 0..max_quant()). Geometric color
   * difference is basically the magnitude of the colour interpreted as 3d vector
   * compared, instead of simple cv::inRange(). For gray scale images,
   * geometric colour difference is not applied.
   * @param const color_t &c
   * @param float_t max_difference=0.1
   * @param bool geometric_color_difference=false
   * @return basic_image
   */
  basic_image in_color_range(const color_t &c, float_t max_difference=0.1,
      bool geometric_color_difference=false)
  {
    double d = max_difference * max_quant();
    if(channels()==1) {
      mat_t m(m_.rows, m_.cols, CV_8U);
      cv::inRange(m_, cv::Scalar(c.r()-d), cv::Scalar(c.r()+d), m);
      return m;
    } else if(!geometric_color_difference) {
      mat_t m(m_.rows, m_.cols, CV_8U);
      cv::inRange(m_,
        cv::Scalar(c.b()-d, c.g()-d, c.r()-d),
        cv::Scalar(c.b()+d, c.g()+d, c.r()+d),
      m);
      return m;
    } else {
      cv::Mat3b m;
      {
        basic_image im(*this);
        im.to_rgb(); // Ensure BGR uint8_t, continuous
        m = im.mat();
      }
      cv::Mat1b r(m.rows, m.cols);
      cv::Mat1b::iterator rit = r.begin();
      for(cv::Mat3b::const_iterator it=m.begin(); it != m.end(); ++it, ++rit) {
        long cb=(*it)[0]-c.b(), cg=(*it)[1]-c.g(), cr=(*it)[2]-c.r();
        *rit = sqrt(cb*cb+cg*cg+cr*cr) <= d ? 255 : 0;
      }
      return r;
    }
  }
  // </editor-fold>
 
public: // Methods modifying the image (and returning a reference to itself))
 
  // <editor-fold desc="clear" defaultstate="collapsed">
  /**
   * Clears the contents and resets all instance variables to defaults.
   * @return basic_image&
   */
  inline basic_image& clear()
  { m_ = cv::Mat(); is_hsv_ = false; file_ = ""; return *this; }
 
  // </editor-fold>
 
  // <editor-fold desc="negate/invert" defaultstate="collapsed">
 
  /**
   * Negate the image ("invert")
   * @return basic_image&
   */
  inline basic_image& negate()
  {
    if(m_.channels() == 1) {
      m_ = max_quant() - m_;
    } else {
      mat_vector_t im;
      cv::split(m_, im);
      for(typename mat_vector_t::iterator it=im.begin(); it!=im.end(); ++it) {
        (*it) = max_quant() - (*it);
      }
      cv::merge(im, m_);
      im.clear();
    }
    return *this;
  }
 
  /**
   * Negate the image ("invert")
   * @return basic_image&
   */
  inline basic_image& invert()
  {
    return negate();
  }
 
  // </editor-fold>
 
  // <editor-fold desc="extend, crop" defaultstate="collapsed">
 
  typedef enum { em_color=0, em_replicate, em_reflect, em_wrap } extend_mode_t;
 
  /**
   * Extend an image area
   * @param const point_t &top_left
   * @param const point_t &bottom_right
   * @param extend_mode_t how=em_color
   * @param const color_t &c=color_t()
   * @return basic_image&
   */
  inline basic_image& extend(const point_t &top_left, const point_t &bottom_right,
      extend_mode_t how=em_color, const color_t &c=color_t())
  {
    int t;
    switch(how) {
      case em_replicate: t=cv::BORDER_REPLICATE; break;
      case em_reflect: t=cv::BORDER_REFLECT; break;
      case em_wrap: t=cv::BORDER_WRAP; break;
      case em_color: default: t=cv::BORDER_CONSTANT;
    }
    cv::copyMakeBorder(m_, m_, top_left.y(), bottom_right.y(), top_left.x(),
        bottom_right.x(), t, (cv::Scalar) c);
    return *this;
  }
 
  /**
   * Extend an image area
   * @param crd_t px
   * @param extend_mode_t how=em_color
   * @param const color_t &c=color_t()
   * @return basic_image&
   */
  inline basic_image& extend(crd_t px, extend_mode_t how=em_color,
      const color_t &c=color_t())
  { return extend(point_t(px,px), point_t(px,px), how, c); }
 
 
  /**
   * Extend to a specified width and height. If `centered==true`, border at top,
   * left, bottom, right is added; otherwise only bottom and right.
   * @param crd_t width
   * @param crd_t height
   * @param bool centered=false
   * @param extend_mode_t how=em_color
   * @param const color_t &c=color_t()
   * @return basic_image&
   */
  inline basic_image& extend(crd_t width, crd_t height, bool centered=false,
      extend_mode_t how=em_color, const color_t &c=color_t())
  {
    if(width<this->width() && height<this->height()) return *this;
    crd_t dw = width < this->width() ? 0 : width-(this->width());
    crd_t dh = height < this->height() ? 0 : height-(this->height());
    crd_t t, b, l, r;
    if(centered) {
      l = r = dw/2; r += (l+r < dw) ? 1 : 0; // rounding
      t = b = dh/2; b += (t+b < dh) ? 1 : 0;
    } else {
      t = l = 0; r = dw; b = dh;
    }
    dbg2("t=" << t << ", l=" << l << ", b=" << b << ", r=" << r);
    return extend(point_t(t,l), point_t(b,r), how, c);
  }
 
  /**
   * Crop the image to the specified region
   * @param const rect_t& region
   * @return basic_image&
   */
  inline basic_image& crop(const rect_t& region)
  {
    crd_t x0, y0, x1, y1;
    if(region.width()<1 || region.height()<1) { clear(); return *this; }
    x0 = region.tl().x(); y0 = region.tl().y();
    x1 = region.br().x(); y1 = region.br().y();
    x0 = x0<0 ? 0 : x0;
    y0 = y0<0 ? 0 : y0;
    x1 = x1>=width() ? width()-1 : x1;
    y1 = y1>=height() ? height()-1 : y1;
    rect_t r(x0, y0, x1, y1);
    if(r.width() >= width() && r.height() > height()) return *this;
    m_ = m_((cv::Rect)r).clone(); // Re-assign to ensure continuous mat.
    return *this;
  }
 
  /**
   * Crop the image to the specified width/height from the the bottom-right
   * corner.
   * @param crd_t width
   * @param crd_t height
   * @return basic_image&
   */
  inline basic_image& crop(crd_t width, crd_t height)
  { return crop(rect_t(0, 0, width, height)); }
 
  // </editor-fold>
 
  // <editor-fold desc="resample, stretch, scale, resize" defaultstate="collapsed">
 
  /**
   * Resamples the image, doubles the width and height.
   * @return basic_image&
   */
  inline basic_image& resample_up()
  { cv::pyrUp(m_, m_); return *this; }
 
  /**
   * Resamples the image, halves the width and height.
   * @return basic_image&
   */
  inline basic_image& resample_down()
  { cv::pyrDown(m_, m_); return *this; }
 
 
  /**
   * Resizing interpolation methods
   * @enum resize_interpolation_t
   */
  typedef enum {
    resz_bilinear = cv::INTER_LINEAR,
    resz_nearest = cv::INTER_NEAREST,
    resz_cubic = cv::INTER_CUBIC,
    resz_area = cv::INTER_AREA,
    resz_laczos4 = cv::INTER_LANCZOS4
  } resize_interpolation_t;
 
  /**
   * Resizes the image to (width)x(height) without keeping the aspect ratio.
   * @param crd_t width
   * @param crd_t height
   * @param resize_interpolation_t interpolation
   * @return basic_image&
   */
  inline basic_image& resize_stretch(crd_t width, crd_t height, resize_interpolation_t
    interpolation=resz_area)
  {
    if(width<=0 || height<=0) { m_ = cv::Mat(0,0,m_.type()); return *this; }
    cv::resize(m_, m_, cv::Size(width, height), 0, 0, (int)interpolation);
    return *this;
  }
 
  /**
   * Resizes the image using a scale factor (>0). The factor refers to the edge lengths.
   * @param float_t factor
   * @return basic_image&
   */
  inline basic_image& resize_scale(float_t factor)
  {
    if(factor<=0) return *this;
    resize_stretch(::round(factor*width()), ::round(factor*height()),
        factor<1 ? resz_area : resz_nearest);
    return *this;
  }
 
  /**
   * Resizes the image to a maximum height/width. Keeps aspect ratio. Scales
   * up and down.
   * @param crd_t max_width
   * @param crd_t max_height
   * @return basic_image&
   */
  inline basic_image& resize(crd_t max_width, crd_t max_height)
  {
    float_t fw=(float_t)max_width/width(), fh=(float_t)max_height/height();
    if(fw < fh) {
      return resize_stretch(max_width, fw*height());
    } else {
      return resize_stretch(fh*width(), max_height);
    }
  }
 
  // </editor-fold>
 
  // <editor-fold desc="level, normalize, histogram_equalize" defaultstate="collapsed">
  /**
   * Linear stretch the colour value from min->max ---> 0->max_quant()
   * @return basic_image&
   */
  inline basic_image& level()
  {
    if(m_.channels() == 1) {
      double min, max;
      cv::minMaxLoc(m_, &min, &max);
      if(max > min) { m_ -= min; m_ *= (max_quant() / (max-min)); }
    } else {
      mat_vector_t im;
      cv::split(m_, im);
      for(typename mat_vector_t::iterator it=im.begin(); it!=im.end(); ++it) {
        double min, max;
        cv::minMaxLoc((*it), &min, &max);
        if(max > min) { (*it) -= min; (*it) *= (max_quant() / (max-min)); }
      }
      cv::merge(im, m_);
      im.clear();
    }
    return *this;
  }
 
  /**
   * Linear stretch the colour value from min->max ---> 0->max_quant(),
   * where the image colour range is cropped to min_0_1, max_0_1.
   * @param double min_0_1
   * @param double max_0_1
   * @return basic_image&
   */
  basic_image& level(double min_0_1, double max_0_1)
  {
    min_0_1 = min_0_1 > 1 ? 1 : min_0_1 < 0 ? 0 : min_0_1;
    max_0_1 = max_0_1 > 1 ? 1 : max_0_1 < 0 ? 0 : max_0_1;
    if(channels() == 1) {
      m_.convertTo(m_, CV_8U);
      cv::threshold(m_, m_, min_0_1*max_quant(), max_quant(), cv::THRESH_TOZERO);
      cv::threshold(m_, m_, max_0_1*max_quant(), max_quant(), cv::THRESH_TRUNC);
      double min, max;
      cv::minMaxLoc(m_, &min, &max);
      if(max > min) { m_ -= min; m_ *= (max_quant() / (max-min)); }
    } else {
      mat_vector_t im;
      cv::split(m_, im);
      for(typename mat_vector_t::iterator it=im.begin(); it!=im.end(); ++it) {
        (*it).convertTo((*it), CV_8U);
        cv::threshold((*it), (*it), min_0_1*max_quant(), max_quant(), cv::THRESH_TOZERO);
        cv::threshold((*it), (*it), max_0_1*max_quant(), max_quant(), cv::THRESH_TRUNC);
        double min, max;
        cv::minMaxLoc((*it), &min, &max);
        if(max > min) { (*it) -= min; (*it) *= (max_quant() / (max-min)); }
      }
      cv::merge(im, m_);
      im.clear();
    }
    return *this;
  }
 
  /**
   * Normalise image to values between min (0 to 1.0) and max (0 to 1.0).
   * @param double min_0_1
   * @param double max_0_1
   * @return basic_image&
   */
  inline basic_image& normalize(double min_0_1, double max_0_1)
  {
    min_0_1 = min_0_1 > 1 ? 1 : min_0_1 < 0 ? 0 : min_0_1;
    max_0_1 = max_0_1 > 1 ? 1 : max_0_1 < 0 ? 0 : max_0_1;
    cv::normalize(m_, m_, min_0_1*max_quant(), max_0_1*max_quant(), cv::NORM_MINMAX);
    return *this;
  }
 
  /**
   * Equalises the image histogram (for each channel independently)
   * @return basic_image&
   */
  inline basic_image& histogram_equalize()
  {
    if(channels()==1) {
      cv::equalizeHist(m_, m_);
    } else {
      mat_vector_t chs = split_channels();
      for(unsigned i=0; i<chs.size(); ++i)
        cv::equalizeHist(chs[i], chs[i]);
      cv::merge(chs, m_);
    }
    return *this;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="threshold, threshold_adaptive" defaultstate="collapsed">
  /**
   * Sets all pixels with a value smaller than `value_0_1*max_quant()` to 0,
   * all other pixels to max_quant().
   * @param double value_0_1
   * @return basic_image&
   */
  inline basic_image& threshold(double value_0_1=.5)
  {
    value_0_1 = value_0_1 > 1 ? 1 : value_0_1 < 0 ? 0 : value_0_1;
    to_grayscale();
    cv::threshold(m_, m_, value_0_1*max_quant(), max_quant(), cv::THRESH_BINARY);
    return *this;
  }
 
  /**
   * Sets all pixels with a value smaller than the corresponding component of the colour
   * to black.
   * @param double value_0_1
   * @return basic_image&
   */
  inline basic_image& threshold(const color_t& c)
  {
    if(is_gray() || (c[0]==c[1] && c[0]==c[2])) {
      int cc = c[0];
      if(cc < c[1]) cc = c[1];
      if(cc < c[2]) cc = c[2];
      if(cc >= max_quant()) cc = max_quant()-1;
      cv::threshold(m_, m_, cc, max_quant(), cv::THRESH_BINARY);
    } else {
      mat_vector_t chs = split_channels();
      if(chs.size() > 3) chs.resize(3);
      for(typename mat_vector_t::size_type i=0; i<chs.size(); ++i) {
        int cc = c[i] >= max_quant() ? max_quant()-1 : c[i];
        cv::threshold(chs[i], chs[i], cc, max_quant(), cv::THRESH_BINARY);
      }
      merge_channels(chs);
    }
    return *this;
  }
 
  /**
   * Adaptive threshold
   * @param crd_t radius=0
   * @param bool gauss=false
   * @return basic_image&
   */
  inline basic_image& threshold_adaptive(crd_t radius=0, bool gauss=false)
  {
    to_grayscale();
    if(radius<2) radius = 2;
    cv::adaptiveThreshold(m_, m_, max_quant(), gauss ? cv::ADAPTIVE_THRESH_GAUSSIAN_C :
      cv::ADAPTIVE_THRESH_MEAN_C, cv::THRESH_BINARY, 2*radius+1, 0);
    return *this;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="dilate, erode" defaultstate="collapsed">
  /**
   * Dilates the image
   * @param unsigned radius
   * @return basic_image&
   */
  inline basic_image& dilate(unsigned radius=1)
  {
    cv::dilate(m_, m_, cv::getStructuringElement(cv::MORPH_ELLIPSE,
      cv::Size(2*radius+1, 2*radius+1), cv::Point(radius, radius)
    ));
    return *this;
  }
 
  /**
   * Erodes the image
   * @param unsigned radius
   * @return basic_image&
   */
  inline basic_image& erode(unsigned radius=1)
  {
    cv::erode(m_, m_, cv::getStructuringElement(cv::MORPH_ELLIPSE,
      cv::Size(2*radius+1, 2*radius+1), cv::Point(radius, radius)
    ));
    return *this;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="blur" defaultstate="collapsed">
 
  /**
   * blur() methods:
   */
  typedef enum { blur_box=0, blur_gauss, blur_median } blur_type;
 
  /**
   * Blur with given "radius" (kernel size 2*dist+1).
   * @param int kernel_radius=1
   * @return basic_image&
   */
  inline basic_image& blur(int kernel_radius=1, blur_type how=blur_gauss)
  {
    if((kernel_radius=kernel_radius > 32 ? 32 : kernel_radius) > 0) {
      if(how == blur_gauss) {
        cv::GaussianBlur(m_, m_, cv::Size(2*kernel_radius+1, 2*kernel_radius+1), 0, 0 );
      } else if(how == blur_median) {
        cv::medianBlur(m_, m_, 2*kernel_radius+1);
      } else {
        cv::blur(m_, m_, cv::Size(2*kernel_radius+1, 2*kernel_radius+1));
      }
    }
    return *this;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="sobel, scharr, laplacian, gradient" defaultstate="collapsed">
 
  /**
   * Applies Sobel spatial image derivative. Returns the absolute values of
   * each (pixel) location derivative (values are 0 to something >0).
   * The "kernel_radius" expands to kernel size 2*kernel_radius+1.
   * Note: dx, dy are the filter orders in x/y, valid values
   * 0(=component off) to kernel_size. On scale_to_max_quantum scales==true
   * the values are shifted half the quantum range (127 for 8 bit), so that
   * negative gradients are <128, positive gradient values >=128;
   * @param int dx=1
   * @param int dy=1
   * @param int kernel_radius=0
   * @return basic_image
   */
  inline basic_image& sobel(int dx=1, int dy=1, int kernel_radius=0)
  {
    kernel_radius = kernel_radius<0 ? 0:kernel_radius;
    kernel_radius = 2*kernel_radius+1;
    dx=dx<0?0: (dx>kernel_radius? kernel_radius:dx);
    dy=dy<0?0: (dy>kernel_radius? kernel_radius:dy);
    mat_t m;
    cv::Sobel(m_, m, CV_32F, dx, dy, kernel_radius, 1, 0, cv::BORDER_DEFAULT);
    m = abs(m);
    double mi, ma;
    cv::minMaxLoc(m, &mi, &ma);
    m *= (double)max_quant()/ma;
    m_ = m;
    return *this;
  }
 
  /**
   * Applies Scharr filter. Returns the absolute values of each (pixel) location
   * derivative (values are normed 0 to max_quant()). The data type of each pixel
   * is float32.
   * Note: dx, dy are the filter orders in x/y, valid values
   * 0(=component off) to kernel_size. On scale_to_max_quantum scales==true
   * the values are shifted half the quantum range (127 for 8 bit), so that
   * negative gradients are <128, positive gradient values >=128;
   * @param int dx
   * @param int dy
   * @return basic_image&
   */
  inline basic_image& scharr(int dx, int dy)
  {
    dx=dx<0?0: (dx>1? 1:dx);
    dy=dy<0?0: (dy>1? 1:dy);
    mat_t m;
    cv::Scharr(m_, m, CV_32F, dx, dy, 1, 0, cv::BORDER_DEFAULT);
    m = abs(m);
    double mi, ma;
    cv::minMaxLoc(m, &mi, &ma);
    m *= (double)max_quant()/ma;
    m_ = m;
    return *this;
  }
 
  /**
   * Laplacian ("second derivative"). The "kernel_radius" expands to kernel size
   * 2*kernel_radius+1.
   * @param int kernel_radius=0
   * @return basic_image&
   */
  inline basic_image& laplacian(int kernel_radius=0)
  {
    kernel_radius=kernel_radius<0 ? 0:kernel_radius;
    mat_t m0; mat().convertTo(m0, CV_32F);
    mat_t m(m0.rows, m0.cols, m0.type());
    cv::Laplacian(m0, m, CV_32F, 2*kernel_radius+1, 1, 0, cv::BORDER_DEFAULT);
    m = abs(m);
    double mi, ma;
    cv::minMaxLoc(m, &mi, &ma);
    m *= (double)max_quant()/ma;
    m_ = m;
    return *this;
  }
 
  /**
   * Gradient calculation from Sobel in x and y. Returns ("geometric") magnitude
   * of x/y values (normed to max_quant()). The data type of the image object
   * changes to float_t. The "kernel_radius" expands to kernel size
   * 2*kernel_radius+1.
   *
   * @param unsigned kernel_radius=0
   * @return basic_image&
   */
  inline basic_image& gradient(int kernel_radius=0)
  {
    if(channels()>1) to_grayscale();
    cv::Mat1d xm, ym;
    {
      cv::Sobel(m_, xm, CV_64F, 1,0, 2*kernel_radius+1, 1,0, cv::BORDER_DEFAULT);
      cv::Sobel(m_, ym, CV_64F, 0,1, 2*kernel_radius+1, 1,0, cv::BORDER_DEFAULT);
    }
    if(ym.size != xm.size) {
      dbg("Sobel x/y derivative matrices SIZE MISSMATCH");
    } else {
      cv::Mat1d mags(xm.rows, xm.cols);
      cv::magnitude(xm,ym,mags);
      double mi, ma;
      mags = abs(mags);
      cv::minMaxLoc(mags, &mi, &ma);
      if(ma!=0) mags *= (double)max_quant()/ma;
      mags.convertTo(m_, sizeof(float_t)==8 ? CV_64F : CV_32F);
      //dbg("gradient(): max magnitude=" << ma);
    }
    return *this;
  }
  // </editor-fold>
 
  // <editor-fold desc="edge/corner detection" defaultstate="collapsed">
 
  /**
   * Canny edge filter. The `kernel_radius` expands to 2*kernel_radius+1.
   * @param double thres1=100
   * @param double thres2=120
   * @param int kernel_radius=1
   * @return basic_image&
   */
  inline basic_image& edges_canny(crd_t thres1=100, crd_t thres2=120, int kernel_radius=1)
  { cv::Canny(m_, m_, thres1, thres2, 2*kernel_radius+1, true); return *this;}
 
  /**
   * Harris corner detector. The `kernel_radius` expands to 2*kernel_radius+1.
   * @param k=0.01
   * @param block_size=3
   * @param kernel_radius=1
   * @return basic_image&
   */
  inline basic_image& corners_harris(double k=0.01, int block_size=3, int kernel_radius=1)
  {
    if(channels() > 1) to_grayscale();
    mat_t m = mat_t::zeros(m_.size(), CV_32FC1);
    cv::cornerHarris(m_, m, block_size, 2*kernel_radius+1, k, cv::BORDER_DEFAULT );
    cv::normalize(m, m, 0, max_quant(), cv::NORM_MINMAX, CV_32FC1, mat_t());
    cv::convertScaleAbs(m, m_);
    return *this;
  }
  // </editor-fold>
 
public:  // Methods returning non-image data types
 
  // <editor-fold desc="contours" defaultstate="collapsed">
 
  typedef enum {
    cfa_code=0, cfa_none=1, cfa_simple=2, cfa_l1=3, cfa_kcos=4, cfa_link_runs=5
  } contour_chain_approx_t;
 
  /**
   * Find contours. Every nonzero (nonblack) pixel is used. Colour images are
   * (copy) converted to gray scale.
   * @param bool external_only=false
   * @param contour_chain_approx_t chain_approx=cf_simple
   * @return contour_list_t
   */
  inline contour_list_t contours(bool external_only=false,
          contour_chain_approx_t chain_approx=cfa_simple) const
  {
    contour_list_t contours;
    std::vector<std::vector<cv::Point2i> > cvcontours;
    {
      std::vector<cv::Vec4i> hierarchy;
      basic_image im(m_);
      if(im.channels()>1) im.to_grayscale();
      cv::findContours(im.mat(), cvcontours, hierarchy, external_only ?
        CV_RETR_EXTERNAL : CV_RETR_LIST, chain_approx);
    }
    while(!cvcontours.empty()) {
      contours.push_back((contour_t)cvcontours.back());
      cvcontours.pop_back();
    }
    return contours;
  }
 
  /**
   * Find contours. Every nonzero (nonblack) pixel is used. Colour images are
   * (copy) converted to gray scale. Contours are organised in external boundaries
   * and holes. "Holes in holes" are treated as external structures again, and
   * so on (even contour levels are external, odd levels are holes).
   * @param contour_chain_approx_t chain_approx=cf_simple
   * @return contour_ccomp_list_t
   */
  inline contour_ccomp_list_t contours_ccomp(
          contour_chain_approx_t chain_approx=cfa_simple) const
  {
    contour_ccomp_list_t contours;
    std::vector<std::vector<cv::Point2i> > cvcontours;
    std::vector<cv::Vec4i> cvhierarchy;
    {
      basic_image im(m_);
      if(im.channels()>1) im.to_grayscale();
      cv::findContours(im.mat(), cvcontours, cvhierarchy, CV_RETR_CCOMP, chain_approx);
    }
    while(!cvcontours.empty()) {
      if(cvhierarchy.back()[3] < 0) {
        contours.outer.push_back((contour_t)cvcontours.back());
      } else {
        contours.inner.push_back((contour_t)cvcontours.back());
      }
      cvcontours.pop_back();
      cvhierarchy.pop_back();
    }
    return contours;
  }
 
  /**
   * Finds contours and returns them as tree structure, where inner contours
   * the child elements. The root node contour is always the whole image (t.m.
   * a rectangular contour with 4 points: (0,0,image width, image height).
   * @param contour_chain_approx_t chain_approx=cfa_simple
   * @return contour_nested_list_t
   */
  contour_nested_list_t contours_tree(
    contour_chain_approx_t chain_approx=cfa_simple) const
  {
    if(!width() || !height()) return (contour_t(rect_t(point_t(0,0),point_t(0,0))));
    contour_nested_list_t contours(contour_t(rect_t(0,0,width()-1,height()-1)));
    {
      std::vector<contour_nested_list_t> nodes; // Nodes (contents);
      std::vector<int> parent_indices, parents;
      {
        std::vector<std::vector<cv::Point2i> > cvcontours;
        std::vector<cv::Vec4i> cvhierarchy;
        {
          basic_image im(m_);
          if(im.channels()>1) im.to_grayscale();
          cv::findContours(im.mat(), cvcontours, cvhierarchy, CV_RETR_TREE, chain_approx);
        }
        std::map<int,int> avail_parents;
        for(unsigned i=0; i<cvcontours.size(); ++i) {
          unsigned parent = cvhierarchy[i][3];
          nodes.push_back(contour_nested_list_t(cvcontours[i]));
          parent_indices.push_back(parent);
          avail_parents[parent] = 0;
        }
        for(typename std::map<int,int>::iterator it=avail_parents.begin();
                it!=avail_parents.end(); ++it) {
          parents.push_back(it->first);
        }
        std::sort(parents.begin(), parents.end());
        if(parents.front()==-1) parents.erase(parents.begin());
      } while(0);
 
      // --> Rollup parents from back to front
      while(!parents.empty()) {
        int pi = parents.back();
        parents.pop_back();
        for(unsigned i=0; i<parent_indices.size(); ++i) {
          if(parent_indices[i]==pi) {
            nodes[pi].children_.push_back(nodes[i]); // push copy
            nodes[pi].children_.back().parent_ = &(nodes[pi]);
            nodes[i].clear(); // save mem
          }
        }
      }
      for(unsigned i=0; i<parent_indices.size(); ++i) {
        if(parent_indices[i] < 0) {
          contours.children_.push_back(nodes[i]);
          contours.children_.back().parent_ = &contours;
          nodes[i].clear();
        }
      }
    }
    return contours;
  }
 
  // </editor-fold>
 
  // <editor-fold desc="hough lines/circles" defaultstate="collapsed">
  /**
   * Hough line transform
   * @param double acc_vote_threshold=1
   * @param double min_line_length=10
   * @param double max_line_point_gap=2
   * @param double acc_angle_resolution_deg=1.0
   * @param double acc_distance_resolution_px=1
   * @return line_list_t (vector of line_t)
   */
  inline line_list_t hough_lines(
    double acc_vote_threshold=1,
    double min_line_length=10, double max_line_point_gap=2,
    double acc_angle_resolution_deg=1.0, double acc_distance_resolution_px=1
  ){
    std::vector<cv::Vec4i> hlines;
    line_list_t lines;
    cv::HoughLinesP(m_, hlines,
      acc_distance_resolution_px, acc_angle_resolution_deg*M_PI/360,
      acc_vote_threshold, min_line_length, max_line_point_gap
    );
    for(size_t i=0; i<hlines.size(); ++i) {
      const cv::Vec4i &l = hlines[i];
      lines.push_back(line_t(point_t(l[0],l[1]), point_t(l[2],l[3])));
    }
    return lines;
  }
 
  /**
   * Hough circle transform
   * Note: min_centre_distance==-1 --> max_radius
   * @param double max_radius=10
   * @param double min_radius=5
   * @param double canny_threshold=50
   * @param acc_vote_threshold=50
   * @param min_centre_distance=-1
   * @param relative_acc_resolution=1
   * @return circle_list_t (vector of circle_t)
   */
  inline circle_list_t hough_circles(
    double max_radius=10,
    double min_radius=5,
    double canny_threshold=50,
    double acc_vote_threshold=50,
    double min_centre_distance=-1,
    double relative_acc_resolution=1
  ){
    circle_list_t circles;
    std::vector<cv::Vec3f> hc;
    int res = relative_acc_resolution < 2 ? 1 : relative_acc_resolution;
    if(acc_vote_threshold < 1) acc_vote_threshold = 1;
    if(canny_threshold < 1) canny_threshold = 1;
    double dist = min_centre_distance <= 0 ? max_radius : min_centre_distance;
    cv::HoughCircles(
      m_, hc, CV_HOUGH_GRADIENT, res, dist,
      canny_threshold, acc_vote_threshold, min_radius, max_radius
    );
    for(size_t i=0; i<hc.size(); ++i) {
      const cv::Vec3f &c = hc[i];
      circles.push_back(circle_t(c[0],c[1],c[2]));
    }
    return circles;
  }
  // </editor-fold>
 
  // <editor-fold desc="histogram" defaultstate="collapsed">
  /**
   * @see basic_histogram<>
   * @param int num_bins=0
   * @param const basic_image& mask=basic_image()
   * @return histogram_t
   */
  histogram_t histogram(int num_bins=0, const basic_image& mask=basic_image()) const
  { return histogram_t(*this, num_bins, mask); }
 
  // </editor-fold>
 
public: // Drawing
 
  // <editor-fold desc="draw, fill, floodfill, mask" defaultstate="collapsed">
 
  /**
   * Draw a point with defined size.
   * @param const point_t &point
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const point_t &point, const color_t &c=color_t(), double size=1)
  {
    if(size < 2) {
      color_at(point.x(), point.y(), c);
    } else {
      cv::circle(m_, (cv::Point)point, 1, (cv::Scalar) c, size);
    }
    return *this;
  }
 
  /**
   * Draw a line with defined width.
   * @param const line_t &line
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const line_t &line, const color_t &c=color_t(), double size=1)
  { cv::line(m_, (cv::Point)line.p0(), (cv::Point)line.p1(), (cv::Scalar) c, size);
      return *this; }
 
  /**
   * Draw a circle outline with defined line width.
   * @param const circle_t &circle
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const circle_t &circle, const color_t &c=color_t(), double size=1)
  {
    if(circle.r() <= 0 || circle.x() <=0 || circle.y() <=0) return *this;
    cv::circle(m_, (cv::Point)circle, circle.r(), (cv::Scalar) c, size); return *this;
  }
 
  /**
   * Draw an ellipse outline with defined line width.
   * @param const ellipse_t &e
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const ellipse_t &e, const color_t &c=color_t(), double size=1)
  { cv::ellipse(m_, (cv::RotatedRect)e, (cv::Scalar) c, size); return *this; }
 
  /**
   * Draw an upright rectangle outline
   * @param const rect_t &r
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const rect_t &r, const color_t &c=color_t(), double size=1)
  { cv::rectangle(m_, (cv::Rect)r, (cv::Scalar) c, size); return *this; }
 
  /**
   * Draw a rotated rectangle outline
   * @param const rotated_rect_t &r
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const rotated_rect_t &r, const color_t& c=color_t(), double size=1)
  {
    if(r.size().x() && r.size().y() && size >= 1) {
      cv::polylines(m_, std::vector<std::vector<cv::Point2i> >(1, r), true,
          (cv::Scalar) c, (int)size);
    }
    return *this;
  }
 
  /**
   * Draw a polynom outline.
   * @param const polynom_t &p
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const polynom_t &p, const color_t &c=color_t(), double size=1)
  {
    if(!p.empty()) {
      cv::polylines(m_, std::vector<std::vector<cv::Point2i> >(1,p), p.closed(),
          (cv::Scalar) c, (int)size);
    }
    return *this;
  }
 
  /**
   * Draw a contour outline
   * @param const contour_t &p
   * @param const color_t &c=color_t()
   * @param double size=1
   * @return basic_image& (*this)
   */
  inline basic_image& draw(const contour_t &p, color_t c=color_t(), double size=1)
  {
    if(!p.empty()) {
      cv::polylines(m_, std::vector<std::vector<cv::Point2i> >(1,p), true,
      (cv::Scalar) c, (int)size);
    }
    return *this;
  }
 
  /**
   * Draws a channel of the histogram
   * @param const histogram_t &hist
   * @param int channel
   * @param const img_t::rect_t &where=img_t::rect_t()
   * @param color_t curve_color = white
   * @param color_t background_color = black
   * @param color_t border_color = gray
   * @param int grid_div_x = 4
   * @param int grid_div_y = 4
   * @param color_t grid_color = gray
   * @param int line_width = 0
   * @return basic_image&
   */
  inline basic_image& draw(const histogram_t &hist, int channel,
      const rect_t &where=rect_t(), color_t curve_color=color_t(255),
      color_t background_color=color_t(), color_t border_color=color_t(80),
      int grid_div_x=4, int grid_div_y=4, color_t grid_color=color_t(80), int line_width=0
      )
  {
    hist.draw_channel(*this, channel, where, curve_color, background_color, border_color,
      grid_div_x, grid_div_y, grid_color, line_width);
    return *this;
  }
 
  /**
   * Draws an image into another one the image is scaled to the rect size.
   * @param const basic_image& im
   * @param rect_t where
   * @return basic_image&
   */
  inline basic_image& draw(const basic_image& im, rect_t where)
  {
    if(where.width()<1 || where.height()<1 || im.empty()) return *this;
    basic_image iim(im);
    iim.resize_stretch(where.width(), where.height());
    if(where.br().x()>=width() || where.br().y()>=height()) {
      where = rect_t(point_t(where.p()), point_t(width()-1, height()-1));
      iim.crop(where.width(), where.height());
      dbg2("draw(): Modified (rect_t) where=" << where);
    }
    cv::Mat mm = m_((cv::Rect) where);
    iim.mat().copyTo(mm);
    return *this;
  }
 
  /**
   * Draws an image into another one with mask.
   * @param const basic_image& im
   * @param const basic_image& mask
   * @return basic_image&
   */
  inline basic_image& draw(const basic_image& im, const basic_image& mask)
  {
    if(color_space() != im.color_space()) {
      basic_image im2(im);
      im2.color_space(color_space());
      im2.mat().copyTo(m_, mask.mat());
    } else {
      im.mat().copyTo(m_, mask.mat());
    }
    return *this;
  }
 
  /**
   * Draws the histogram
   * @param const histogram_t &hist
   * @param const img_t::rect_t &where=img_t::rect_t()
   * @return basic_image&
   */
  inline basic_image& draw(const histogram_t &hist, const rect_t &where)
  { hist.draw(*this, where); return *this; }
 
  /**
   * Fill the whole image with a defined colour.
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const color_t &c)
  { m_ = (cv::Scalar)c; return *this; }
 
  /**
   * Fill am area defined by a CV point vector
   * @param const std::vector<cv::Point2i> &p
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const std::vector<cv::Point2i> &p, const color_t &c=color_t())
  { if(p.empty()) return *this;
    std::vector<std::vector<cv::Point2i> > v; v.push_back((std::vector<cv::Point2i>)p);
    cv::fillPoly(m_, v,(cv::Scalar) c); return *this; }
 
  /**
   * Fill a circle area
   * @param const circle_t &circle
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const circle_t &cr, const color_t &c=color_t())
  { if(cr.r()>0) cv::circle(m_, (cv::Point)cr, cr.r(), (cv::Scalar) c, ls_filled);
      return *this; }
 
  /**
   * Fill an ellipse area
   * @param const ellipse_t &e
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const ellipse_t &e, const color_t &c=color_t())
  { cv::ellipse(m_, (cv::RotatedRect)e, (cv::Scalar) c, ls_filled); return *this; }
 
  /**
   * Fill an upright rectangle area
   * @param const rect_t &r
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const rect_t &r, const color_t &c=color_t())
  { cv::rectangle(m_, (cv::Rect)r, (cv::Scalar) c, ls_filled); return *this; }
 
  /**
   * Fill the area of a rotated rectangle
   * @param const rotated_rect_t &r
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const rotated_rect_t &r, const color_t &c=color_t())
  { return fill((std::vector<cv::Point2i>)r, c); }
 
  /**
   * Fill the area of a polynom
   * @param const polynom_t &p
   * @param const color_t &c=color_t()
   * @return basic_image& (*this)
   */
  inline basic_image& fill(const polynom_t &p, const color_t &c=color_t())
  { return fill((std::vector<cv::Point2i>)p, c); }
 
  /**
   * Flood fill from seed point with defined colour and tolerance. The tolerance
   * of type color_t can be specified as integer as well and will be implicitly
   * converted to the corresponding gray scale colour.
   * @param const point_t& seed_point
   * @param const color_t& c
   * @param const color_t& tolerance
   * @param bool difference_from_seed_point=false
   * @return basic_image&
   */
  inline basic_image& floodfill(const point_t &seed_point, const color_t &c,
      const color_t &tolerance, bool difference_from_seed_point=false)
  {
    int flags = 4|(255<<8)|(difference_from_seed_point ? cv::FLOODFILL_FIXED_RANGE:0);
    cv::Mat mask;
    cv::floodFill(m_, mask, (cv::Point) seed_point, (cv::Scalar)c , 0,
        (cv::Scalar)tolerance, (cv::Scalar)tolerance, flags);
    return *this;
  }
 
  /**
   * Flood fill from seed point with defined colour and tolerance. The tolerance
   * of type color_t can be specified as integer as well and will be implicitly
   * converted to the corresponding gray scale colour.
   * @param const point_t& seed_point
   * @param const color_t& c
   * @param const color_t& tolerance
   * @param bool difference_from_seed_point
   * @param basic_image& mask
   * @return basic_image&
   */
  inline basic_image& floodfill(const point_t &seed_point, const color_t &c,
      const color_t &tolerance, bool difference_from_seed_point, basic_image& mask)
  {
    int flags = 4|(255<<8)|(difference_from_seed_point ? cv::FLOODFILL_FIXED_RANGE:0);
    mask.to_grayscale().extend(width()+2, height()+2, true); // ensure type, size
    cv::floodFill(m_, mask.mat(), (cv::Point) seed_point, (cv::Scalar)c , 0,
        (cv::Scalar)tolerance, (cv::Scalar)tolerance, flags);
    mask.crop(1, 1, width(), height());
    return *this;
  }
 
  /**
   * Flood fill from seed point with the colour of the seed point. The tolerance
   * of type color_t can be specified as integer as well and will be implicitly
   * converted to the corresponding gray scale colour.
   * @param const point_t &seed_point
   * @param const color_t &tolerance=0
   * @param bool difference_from_seed_point=false
   * @return basic_image&
   */
  inline basic_image& floodfill(const point_t &seed_point, const color_t &tolerance,
      bool difference_from_seed_point=false)
  {
    color_t c = color_at(seed_point);
    int flags = 4|(255<<8)|(difference_from_seed_point ? cv::FLOODFILL_FIXED_RANGE:0);
    mat_t mask;
    cv::floodFill(m_, mask, (cv::Point) seed_point, (cv::Scalar)c , 0,
        (cv::Scalar)tolerance, (cv::Scalar)tolerance, flags);
    return *this;
  }
 
  /**
   * Returns the pixels that would be modified when flood filling from seed
   * point with defined colour and tolerance. The tolerance
   * of type color_t can be specified as integer as well and will be implicitly
   * converted to the corresponding gray scale colour.
   * @param const point_t& seed_point
   * @param const color_t& c
   * @param const color_t& tolerance=0
   * @param bool difference_from_seed_point=false
   * @return basic_image&
   */
  inline basic_image floodfill_mask(const point_t &seed_point,
      const color_t &tolerance=0, bool difference_from_seed_point=false)
  {
    int flags = 4|(255<<8)|cv::FLOODFILL_MASK_ONLY;
    flags |= difference_from_seed_point ? cv::FLOODFILL_FIXED_RANGE : 0;
    mat_t mask = mat_t::zeros(m_.rows+2, m_.cols+2, CV_8U);
    cv::floodFill(m_, mask, (cv::Point) seed_point, 255, 0, (cv::Scalar)tolerance,
        (cv::Scalar)tolerance, flags);
    return mask(cv::Rect(1,1,m_.cols-2,m_.rows-2)).clone();
  }
 
  /**
   * Masks the image.
   * @param const basic_image & mask
   * @param const color_t& repl_color=color_t::black
   * @return basic_image&
   */
  inline basic_image& mask(const basic_image & mask, const color_t& repl_color=color_t::black)
  {
    basic_image m = mask;
    m.negate();
    mat().setTo((cv::Scalar)repl_color, m.mat());
    return *this;
  }
 
  // </editor-fold>
 
protected:
 
  // <editor-fold desc="instance variables" defaultstate="collapsed">
  mat_t m_;
  std::basic_string<char_t> file_;
  bool is_hsv_;
  // </editor-fold>
 
};
 
// <editor-fold desc="basic_image operator <<" defaultstate="collapsed">
/**
 * Output stream dump
 * @param std::ostream&
 * @param basic_image<crd_t,val_t,float_t>
 * @return std::ostream&
 */
template <typename crd_t, typename val_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_image<crd_t,val_t,float_t> im)
{
  os << "{" << std::endl
     << "  width: " << im.width() << "," << std::endl
     << "  height: " << im.height() << "," << std::endl
     << "  type: \"" << (im.is_gray()?"gray":im.is_hsv()?"hsv":"rgb")<< "\""
     << "," << std::endl
     << "  data_type: \"";
  switch(im.mat().depth()) {
    case CV_8U: os << "uint8"; break;
    case CV_8S : os << "int8"; break;
    case CV_16U: os << "uint16"; break;
    case CV_16S: os << "int16"; break;
    case CV_32S: os << "int32"; break;
    case CV_32F: os << "float32"; break;
    case CV_64F: os << "float64"; break;
    default: os << "unknown";
  }
  os << "\"," << std::endl
     << "  channels: " << im.channels() << "," << std::endl
     << "  min: \"" << im.min() << "\"," << std::endl
     << "  max: \"" << im.max() << "\"," << std::endl
     << "  mean: \"" << im.mean() << "\"," << std::endl
     << "  mat_continuous: " << im.mat().isContinuous() << "," << std::endl
     << "  mat_is_submatrix: " << im.mat().isSubmatrix() << "," << std::endl
     << "  mat_dimensions: " << im.mat().dims << "," << std::endl
     << "  mat_refcount: " << (im.mat().refcount ? *im.mat().refcount : 0) << "," << std::endl
     << "  mat_element_size: " << im.mat().elemSize() << "," << std::endl
     << "  mat_element_cell_size: " << im.mat().elemSize1() << "," << std::endl
     << "  mat_buffer_size: " << im.mat().total() << std::endl
      ;
 
  os << "}" << std::endl;
  return os;
}
// </editor-fold>
 
// <editor-fold desc="basic_histogram" defaultstate="collapsed">
template <typename crd_t, typename val_t, typename float_t>
class basic_histogram
{
public:
 
  typedef basic_image<crd_t,val_t,float_t> img_t;
  typedef typename img_t::rect_t rect_t;
  typedef typename img_t::line_t line_t;
  typedef typename img_t::point_t point_t;
  typedef typename img_t::color_t color_t;
  typedef std::vector<cv::Mat> matv_t;
 
public:
 
  inline basic_histogram() : histograms_(), is_hsv_(false)
  { ; }
 
  inline basic_histogram(const basic_histogram& hist) : histograms_(hist.histograms_),
    is_hsv_(hist.is_hsv_)
  { ; }
 
  /**
   * Directly calculates the histogram. On error, the histogram is empty().
   * @param const img_t &im
   * @param const img_t& mask=img_t()
   * @param unsigned num_bins=0
   */
  inline basic_histogram(const img_t &im, unsigned num_bins=0, const img_t& mask=img_t())
    : histograms_()
  { try{ calculate(im, num_bins, mask); } catch(...) { histograms_.clear(); } }
 
  virtual ~basic_histogram()
  { ; }
 
  inline basic_histogram& operator=(const basic_histogram& h)
  { histograms_=h.histograms_; is_hsv_=h.is_hsv_; return *this; }
 
public:
 
  /**
   * Returns if the histograms is empty.
   * @return bool
   */
  inline bool empty() const throw()
  { return histograms_.empty(); }
 
  /**
   * Returns the number of histogram bins (same for all channels)
   * @return unsigned
   */
  inline unsigned num_bins() const throw()
  { return empty() ? 0 : histograms_.front().rows; }
 
  /**
   * Returns the maximum bin (accumulator) count of a given channel.
   * @param unsigned channel
   * @return int
   */
  inline int max(unsigned channel) const
  {
    if(channel >= histograms_.size()) return 0;
    double mi, ma; cv::minMaxLoc(histograms_[channel], &mi, &ma);
    return ma;
  }
 
  /**
   * Returns the minimum bin (accumulator) count of a given channel.
   * @param unsigned channel
   * @return int
   */
  inline int min(unsigned channel) const
  {
    if(channel >= histograms_.size()) return 0;
    double mi, ma; cv::minMaxLoc(histograms_[channel], &mi, &ma);
    return mi;
  }
 
  /**
   * Returns the raw std::vector<cv::Mat> as const reference.
   * @return matv_t
   */
  inline const matv_t & mats() const throw()
  { return histograms_; }
 
 
  /**
   * Returns the number of histogram bins (same for all channels). This method
   * is used to operate as end index for gray scale histograms combined with
   * operator [].
   * @return unsigned
   */
  inline unsigned size() const throw()
  { return num_bins(); }
 
 
  /**
   * Returns the bin counter of the FIRST CHANNEL at `index` or 0.
   * @param unsigned index
   * @return unsigned
   */
  inline unsigned operator[](unsigned index) const
  { return histograms_.empty() || index >= num_bins() ? 0 :
      (unsigned)histograms_[0].template at<float>(index); }
 
 
public:
 
  /**
   * Calculates the histogram for up to 3 channels. If the specified number of
   * bins is 0, the colour range resolution is assumed (im.max_quant()).
   * @param const img_t &im
   * @param unsigned num_bins=0
   * @param const img_t& mask=img_t()
   * @return basic_histogram&
   */
  basic_histogram& calculate(const img_t &im, unsigned num_bins=0, const img_t& mask=img_t())
  {
    is_hsv_ = im.is_hsv();
    histograms_.clear();
    if(im.empty()) return *this;
    if(num_bins==0) num_bins = im.max_quant()/2+1; // 0xff+1 for 8 bit
    int channels[] = { 0, 1, 2, 3 };
    unsigned n_channels = sizeof(channels) / sizeof(int);
    int n_channel_bins[]  = { (int)num_bins, (int)num_bins, (int)num_bins, (int)num_bins };
    float rng_c0[]={0,is_hsv_?180:256}, rng_c1[]={0,256}, rng_c2[]={0,256}, rng_c3[]={0,256};
    const float* ranges[] = { rng_c0, rng_c1, rng_c2, rng_c3 };
    cv::Mat msk;
    if(!mask.empty()) {
      img_t imsk = mask;
      msk = imsk.to_grayscale().threshold(0.5).mat().clone();
    }
    for(unsigned i=0; i<im.channels() && i<n_channels; ++i) {
      cv::Mat hst;
      cv::calcHist(&im.mat(), 1, channels+i, msk, hst, 1, n_channel_bins+i,
          ranges+i, true, false);
      histograms_.push_back(hst.clone()); // clone: ensure hist not re-referenced
    }
    return *this;
  }
 
  /**
   * Normalises the histogram (all channels if channel<0, selected channel if
   * >=0) to a range from 0 to max_value. This would be the height of the
   * drawing area if the histogram is plotted.
   * @param int max_value
   * @return int channel=-1
   */
  basic_histogram& normalize(int max_value, int channel=-1)
  {
    if(max_value<=1 || channel>=histograms_.size()) return *this;
    int from = 0, to = histograms_.size()-1;
    if(channel >= 0) from = to = channel;
    for(unsigned i=from; i<=to; ++i) {
      cv::normalize(histograms_[i], histograms_[i], 0, max_value, cv::NORM_MINMAX,
        -1, cv::Mat());
    }
    return *this;
  }
 
  /**
   * Draws a channel of the histogram
   * @param img_t& im
   * @param int channel
   * @param const img_t::rect_t &where=img_t::rect_t()
   * @param color_t curve_color = white
   * @param color_t background_color = black
   * @param color_t border_color = gray
   * @param int grid_div_x = 4
   * @param int grid_div_y = 4
   * @param color_t grid_color = gray
   * @param int line_width = 0
   * @return basic_histogram&
   */
  const basic_histogram& draw_channel(img_t& im, int channel, const rect_t &where=rect_t(),
      color_t curve_color=color_t(255), color_t background_color=color_t(),
      color_t border_color=color_t(80), int grid_div_x=4, int grid_div_y=4,
      color_t grid_color=color_t(80), int line_width=0
      ) const
  {
    if(channel<0 || channel>=(int)histograms_.size() || where.width()<5 || where.height()<5) {
      return *this;
    }
    if(line_width<1) {
      line_width = (im.height()>im.width()?im.width():im.height())/500;
      line_width = line_width<0 ? 0 : line_width > 5 ? 5 : line_width;
    }
 
    cv::Mat hst = histograms_[channel].clone();
    int w = where.width()-2; // 2*1px border
    int h = where.height()-2;
    double x_scale = (double)w / num_bins();
    cv::normalize(hst, hst, 0, h, cv::NORM_MINMAX, -1, cv::Mat());
    im.fill(where, background_color);
 
    if(grid_div_x > 1) {
      for(int j=::round((double)w/grid_div_x); j<=w; j+=::round((double)w/grid_div_x)) {
        im.draw(line_t(
          point_t(where.x()+j, where.y()+1),
          point_t(where.x()+j, where.y()+1+h)
        ), grid_color, line_width);
      }
    }
 
    if(grid_div_y > 1) {
      for(int j=::round((double)h/grid_div_y); j<=h; j+=::round((double)h/grid_div_y)) {
        im.draw(line_t(
          point_t(where.x()+1, where.y()+j),
          point_t(where.x()+1+w, where.y()+j)
        ), grid_color, line_width);
      }
    }
 
    im.draw(where, border_color, line_width);
 
    for(int j=0; j<(int)num_bins(); j++) {
      im.fill(rect_t(
          point_t(where.x()+1+(x_scale*(j+0)), where.y()+1+h-hst.at<float>(j)),
          point_t(where.x()+1+(x_scale*(j+1)), where.y()+1+h)
        ), curve_color);
    }
    return *this;
  }
 
  /**
   * Draw a histogram with three graphs (channels left to right)
   * @param img_t& im
   * @param const rect_t &where=rect_t()
   * @return basic_histogram&
   */
  const basic_histogram& draw(img_t& im, const rect_t& where) const
  {
    color_t background_color = color_t();
    color_t border_color = color_t(80);
    int grid_div_x=4;
    int grid_div_y=4;
    color_t grid_color = color_t(80);
    int line_width;
    line_width = (im.height()>im.width()?im.width():im.height())/500;
    line_width = line_width<0 ? 0 : line_width > 5 ? 5 : line_width;
    color_t curve_color = color_t(255);
    int n_ch = histograms_.size();
    if(where.width() < (5+(2*line_width))*n_ch || where.height() < (5+(2*line_width))*1) {
      return *this;
    }
    std::vector<color_t> curve_colors;
    curve_colors.push_back("#ff0000"); // R / H / gray
    curve_colors.push_back("#00ff00"); // G / S
    curve_colors.push_back("#0000ff"); // B / V
    curve_colors.push_back("#ffffff"); // unused (alpha)
    float lw = line_width;
    float gap = 5*lw;
    float rx = where.x() + lw;
    float ry = where.y() + lw;
    float hw = ((float)(where.width()-2*lw)/n_ch)-(n_ch>1?(gap)/(n_ch-1):0)-lw;
    float hh = (float)where.height()-2*lw;
    for(int i=0; i<n_ch; ++i) {
      this->draw_channel(im, i, rect_t(point_t(rx+((hw+gap)*i), ry), hw, hh),
        curve_colors[i], background_color, border_color, grid_div_x, grid_div_y,
        grid_color, line_width
      );
    }
    return *this;
  }
 
protected:
 
  matv_t histograms_;
  bool is_hsv_;
 
};
 
template <typename crd_t, typename val_t, typename float_t>
std::ostream& operator<<(std::ostream& os, basic_histogram<crd_t,val_t,float_t> h)
{
  os << "{" << std::endl
     << "  bins: " << h.num_bins() << "," << std::endl
     << "  channels : " << h.mats().size() << std::endl
     << "}";
  return os;
}
 
// </editor-fold>
 
// <editor-fold desc="basic_camera_capture" defaultstate="collapsed">
template <typename Image_Type, typename Channel_Desc_Type=int>
class basic_camera_capture
{
public:
 
  /**
   * Types
   */
  typedef Image_Type imgage_t;
  typedef Channel_Desc_Type cam_selelct_t;
 
public:
 
  /**
   * Constructors, destructor
   */
  inline basic_camera_capture()
  { ; }
 
  virtual ~basic_camera_capture()
  { close(); }
 
private:
 
  inline basic_camera_capture(const basic_camera_capture & c) : cam_()
  { ; }
 
  inline void operator=(const basic_camera_capture& c)
  { throw "basic_camera_capture: No assignments, please!";  }
 
public:
 
  /**
   * Open the camera channel, returns success
   * @param cam_selelct_t
   * @return bool
   */
  bool open(cam_selelct_t channel)
  { close(); try { return cam_.open(channel) && cam_.isOpened(); } catch(...) { return false; } }
 
  /**
   * Close camera channel
   */
  void close() throw()
  { try { if(cam_.isOpened()) cam_.release(); } catch(...) {;} }
 
  /**
   * Take a snapshot, return success
   * @param imgage_t & img
   * @return bool
   */
  bool snapshot(imgage_t & img)
  {
    if(!cam_.read((::cv::Mat&)img)) { img.clear(); return false; }
    if(!img.mat().isContinuous()) img.mat() = img.mat().clone(); // No fun if that is not given.
    return true;
  }
 
private:
 
  cv::VideoCapture cam_;
};
// </editor-fold>
 
}}}
 
// <editor-fold desc="default specialisations" defaultstate="collapsed">
namespace sw { namespace im {
  typedef detail::basic_image<int, unsigned char, double> cvimage;
  typedef detail::basic_camera_capture<cvimage, int> camera_capture;
  typedef cvimage::exception image_exception;
  typedef cvimage::color_t color;
  typedef cvimage::point_t point2d;
  typedef cvimage::line_t line2d;
  typedef cvimage::line_list_t lines2d;
  typedef cvimage::rect_t rect2d;
  typedef cvimage::circle_t circle2d;
  typedef cvimage::circle_list_t circles2d;
  typedef cvimage::ellipse_t ellipse2d;
  typedef cvimage::rotated_rect_t rotated_rect2d;
  typedef cvimage::polynom_t polynom2d;
  typedef cvimage::contour_t contour2d;
  typedef cvimage::contour_list_t contours2d;
  typedef cvimage::contour_nested_list_t nested_contours2d;
}}
// </editor-fold>
 
// <editor-fold desc="undefs" defaultstate="collapsed">
#undef ex
#undef dbg
#undef dbg2
#endif
// </editor-fold>

Mehr Beispiele

More Examples

/**
 * @ccflags -Wall -O3 -pedantic
 * @ldflags -lm -lopencv_core -lopencv_highgui -lopencv_imgproc
 */
#include <vpu/cvimage.hh>
#include "test.hh"
 
using namespace std;
using namespace cv;
using namespace sw::utest;
 
using sw::im::cvimage;
typedef sw::im::cvimage::color_t color_t;
typedef sw::im::cvimage::point_t point_t;
typedef sw::im::cvimage::line_t line_t;
typedef sw::im::cvimage::rect_t rect_t;
typedef sw::im::cvimage::circle_t circle_t;
typedef sw::im::cvimage::rotated_rect_t rotated_rect_t;
typedef sw::im::cvimage::ellipse_t ellipse_t;
typedef sw::im::cvimage::polynom_t polynom_t;
typedef sw::im::cvimage::contour_t contour_t;
typedef sw::im::cvimage::line_list_t line_list_t;
typedef sw::im::cvimage::circle_list_t circle_list_t;
typedef sw::im::cvimage::contour_list_t contour_list_t;
typedef sw::im::cvimage::contour_ccomp_list_t contour_ccomp_list_t;
typedef sw::im::cvimage::contour_nested_list_t contour_nested_list_t;
typedef sw::im::cvimage::histogram_t histogram_t;
 
 
 
void test_load_save()
{
  cvimage im;
  test_expect_noexcept( im.load("../res/cvimage/im1.png") );
  test_comment( "loaded: " << im.width() << "x" << im.height() );
  if(!test_expect_cond(!im.empty() )) return;
  test_expect( im.max_quant() == 255 );
  test_expect( im.width() == 100 );
  test_expect( im.height() == 75 );
  test_expect( im.area() == im.width()*im.height() );
  test_expect(!im.is_gray() );
  test_expect(!im.is_hsv() );
  test_expect( im.is_rgb() );
  test_expect( im.color_space() == cvimage::csp_rgb );
  test_expect_noexcept( im.to_hsv() );
  test_expect( im.width() == 100 );
  test_expect( im.height() == 75 );
  test_expect( im.area() == im.width()*im.height() );
  test_expect(!im.is_gray() );
  test_expect( im.is_hsv() );
  test_expect(!im.is_rgb() );
  test_expect( im.color_space() == cvimage::csp_hsv );
  test_expect_noexcept( im.to_grayscale() );
  test_expect( im.width() == 100 );
  test_expect( im.height() == 75 );
  test_expect( im.area() == im.width()*im.height() );
  test_expect( im.is_gray() );
  test_expect(!im.is_hsv() );
  test_expect(!im.is_rgb() );
  test_expect( im.color_space() == cvimage::csp_gray );
 
  // Save, reload ...
  test_expect( !im.file().empty() );
  tmp_file f1("im.png"), f2("im.png");
  im.save(f1);
  cvimage im2(f1.path());
  test_expect( im2.file() == f1.path() );
  im2.save(f2);
  test_expect( im2.file() == f1.path() );
}
 
void test_color()
{
  test_comment("Default constructor, ==, !=, !");
  test_expect( color_t() == color_t::black );
  test_expect(!color_t::black );
  test_expect( color_t::black == color_t::black );
  test_expect(!(color_t::black != color_t::black) );
  test_expect( color_t::black != color_t::white );
 
  test_comment("Black components");
  test_expect( color_t::black.r() == 0 );
  test_expect( color_t::black.g() == 0 );
  test_expect( color_t::black.b() == 0 );
  test_expect( color_t::black.a() == 0 );
 
  test_comment("White components");
  test_expect( color_t::white.r() == color_t::max_val() );
  test_expect( color_t::white.g() == color_t::max_val() );
  test_expect( color_t::white.b() == color_t::max_val() );
  test_expect( color_t::white.a() == 0 );
 
  test_comment("to cv::scalar");
  test_expect( ((cv::Scalar)color_t::black)[0] == 0 );
  test_expect( ((cv::Scalar)color_t::white)[0] == 255 );
  test_expect( ((cv::Scalar)color_t::red) == cv::Scalar(0,0,255,0) ); // BGRA
 
  test_comment("from string");
  test_expect( color_t::white == color_t("#ffffff") );
  test_expect( color_t::white == color_t("#fff") );
  test_expect( color_t::white == color_t("ffffff") );
  test_expect( color_t::white == color_t("fff") );
  test_expect( color_t::black == color_t("#000000") );
  test_expect( color_t::black == color_t("#000") );
  test_expect( color_t::black == color_t("000000") );
  test_expect( color_t::black == color_t("000") );
  test_expect( color_t::red == color_t("#f00") );
  test_expect( color_t::green == color_t("#0f0") );
  test_expect( color_t::blue == color_t("#00f") );
 
  test_comment("to string");
  test_expect( (std::string) color_t::black == "#000000" );
  test_expect( (std::string) color_t::white == "#ffffff" );
  test_expect( (std::string) color_t::red   == "#ff0000" );
  test_expect( (std::string) color_t::green == "#00ff00" );
  test_expect( (std::string) color_t::blue  == "#0000ff" );
 
  test_comment("From RGB float components");
  test_expect( color_t::rgb(0, 0, 0) == color_t::black );
  test_expect( color_t::rgb(1, 1, 1) == color_t::white );
  test_expect( color_t::rgb(1, 0, 0) == color_t::red );
  test_expect( color_t::rgb(0, 1, 0) == color_t::green );
  test_expect( color_t::rgb(0, 0, 1) == color_t::blue );
  test_expect(color_t::rgb(.5, .5, .5) == "#7f7f7f" );
 
  test_comment("From HSV float components");
  test_expect( color_t::hsv(0.0 / 3, 1, 1.) == color_t::red );
  test_expect( color_t::hsv(1.0 / 3, 1, 1.) == color_t::green );
  test_expect( color_t::hsv(2.0 / 3, 1, 1.) == color_t::blue );
 
  test_comment("From HSL float components");
  test_expect( color_t::hsl(0.0 / 3, 1, .5) == color_t::red );
  test_expect( color_t::hsl(1.0 / 3, 1, .5) == color_t::green );
  test_expect( color_t::hsl(2.0 / 3, 1, .5) == color_t::blue );
 
  test_comment("Heatmap colors");
  test_expect( color_t::heatmap(0.0) == "#33003f" );
  test_expect( color_t::heatmap(0.1) == "#1a0052" );
  test_expect( color_t::heatmap(0.2) == "#001066" );
  test_expect( color_t::heatmap(0.3) == "#004d79" );
  test_expect( color_t::heatmap(0.4) == "#008c7b" );
  test_expect( color_t::heatmap(0.5) == "#009f3f" );
  test_expect( color_t::heatmap(0.6) == "#0eb200" );
  test_expect( color_t::heatmap(0.7) == "#6ec500" );
  test_expect( color_t::heatmap(0.8) == "#d8d000" );
  test_expect( color_t::heatmap(0.9) == "#eb7100" );
  test_expect( color_t::heatmap(1.0) == "#ff0000" );
}
 
void test_draw()
{
  cvimage im(100, 100, cvimage::csp_rgb);
  if(!test_expect_cond(!im.empty() )) return;
 
  test_expect( !im.mat().empty() );  // cv:Mat access
  test_expect( im.width() == 100 );
  test_expect( im.height() == 100 );
  test_expect( im.area() == 100*100 );
  test_expect( im.channels() == 3 );
  test_expect( im.num_nonzero_pixels() == 0 );
  test_expect( im.is_rgb() );
  test_expect( im.color_at(0,0) == color_t::black );
  test_comment("R" << (int) im.color_at(0,0).r() << ", " <<
               "G" << (int) im.color_at(0,0).g() << ", " <<
               "B" << (int) im.color_at(0,0).b() << ", " <<
               "A" << (int) im.color_at(0,0).a()
              );
  // channel extraction
  test_expect( im.channel_red().num_nonzero_pixels() == 0 );
  test_expect( im.channel_blue().num_nonzero_pixels() == 0 );
  test_expect( im.channel_green().num_nonzero_pixels() == 0 );
 
 // point draw, range check
  test_expect_noexcept( im.draw(point_t(50 , 50), color_t::white, 1) ); // ok
  test_expect_noexcept( im.draw(point_t( -1, -1), color_t::white, 1) ); // no effect
  test_expect_noexcept( im.draw(point_t( -1,  0), color_t::white, 1) ); // no effect
  test_expect_noexcept( im.draw(point_t(  0, -1), color_t::white, 1) ); // no effect
  test_expect_noexcept( im.draw(point_t(100,100), color_t::white, 1) ); // no effect
  test_expect_noexcept( im.draw(point_t(  0,100), color_t::white, 1) ); // no effect
  test_expect_noexcept( im.draw(point_t(100,  0), color_t::white, 1) ); // no effect
  test_expect( im.num_nonzero_pixels() == 1);
 
  // get color, point draw range check
  test_expect( im.color_at( 50, 50) == color_t::white);
  test_expect( im.color_at( -1,  0) == color_t::black);
  test_expect( im.color_at(  0, -1) == color_t::black);
  test_expect( im.color_at(100,  0) == color_t::black);
  test_expect( im.color_at(  0,100) == color_t::black);
  test_expect( !im.color_at( -1,  0));
  test_expect( !im.color_at(  0, -1));
  test_expect( !im.color_at(100,  0));
  test_expect( !im.color_at(  0,100));
  test_expect_noexcept( im.draw(rect_t(25,25,75,75), color_t::blue, 1) );
  test_expect( im.color_at(25, 25) == color_t::blue);
  test_expect_noexcept( im.draw(circle_t(50, 50, 50), color_t::red, 1) );
  test_expect( im.color_at(50,  0) == color_t::red);
  test_expect( im.color_at( 0, 50) == color_t::red);
 
  // fill (1)
  test_expect_noexcept( im.fill("#111") ); // whole image
  test_expect_noexcept( im.fill(circle_t(75, 25, 10), "#333") ); // circle
  test_expect_noexcept( im.fill(rect_t(25,25,10,10), "#555") ); // rect
  test_expect_noexcept( im.fill(ellipse_t(25, 75, 10, 20, -45), "#777") ); // ellipse
  test_expect_noexcept( im.fill(rotated_rect_t(75, 75, 10, 5, 10/*deg*/), "#999") ); // rotated rect
 
  polynom_t ply;
  ply.push_back(point_t(50,50));
  ply.push_back(point_t(50,60));
  ply.push_back(point_t(60,60));
  ply.push_back(point_t(60,70));
  ply.push_back(point_t(70,70));
 
  contour_t cnt;
  cnt.push_back(point_t(50,50));
  cnt.push_back(point_t(50,40));
  cnt.push_back(point_t(40,40));
  cnt.push_back(point_t(40,30));
  cnt.push_back(point_t(30,30));
 
  test_expect_noexcept( im.fill(ply, "#880") );
  test_expect_noexcept( im.draw(ply, "#fff", 1) );
  test_expect_noexcept( im.fill(cnt, "#088") );
  test_expect_noexcept( im.draw(cnt, "#fff", 1) );
 
  test_expect_noexcept( im.draw(line_t(0,0,0,100), "#fff") );
  test_expect_noexcept( im.draw(ellipse_t(50, 50, 100, 50, 45), color_t::red, 1) );
  test_expect_noexcept( im.draw(rotated_rect_t(50, 50, 50, 20, 45/*deg*/), "#777", 1) );
 
  test_expect_noexcept( im.draw(cvimage(im), rect_t(50,50,99,99)) );
  test_expect_noexcept( im.floodfill(point_t(0,0), "#ffff00", color_t(20) /*tolerance*/) );
 
  #if defined (__linux) && defined(DEBUG)
  test_expect_noexcept( im.save("/tmp/testimg.png") );
  #endif
}
 
void test_conv()
{
  cvimage im(50, 50, cvimage::csp_rgb);
  im.fill(color_t::red);
  test_expect( im.color_at(0,0) == color_t::red );
  test_expect( im.color_at(1,1) == color_t::red );
  test_expect_noexcept ( im.to_hsv() );
  test_expect( im.color_at(0,0) == "#ffff00" ); // hue=00, sat=ff, val=ff
  test_expect_noexcept ( im.to_rgb() );
  test_expect( im.color_at(0,0) == "#ff0000" ); // r=ff, g=00, b=00
  test_expect_noexcept ( im.to_grayscale() );
  test_expect( im.color_at(0,0) == "#4c4c4c" ); // 0xffffff / 3
}
 
void test_getters()
{
  cvimage im(100, 50, cvimage::csp_rgb);
  unsigned n = 0;
  for(int c=0; c<im.width(); c+=2) {
    for(int r=0; r<im.height(); r++, ++n) {
      im.draw(point_t(c, r), color_t::white);
      ;
    }
  }
  color_t im_min, im_max, im_mean;
  test_expect( im.num_nonzero_pixels() == n );
  test_expect_noexcept( im.negate() );
  test_expect( im.num_nonzero_pixels() == im.area()-n );
  test_expect_noexcept( im.negate() );
  test_expect_noexcept( im_min  = im.min()  );
  test_expect_noexcept( im_max  = im.max()  );
  test_expect_noexcept( im_mean = im.mean() );
  test_comment("image min  = " << im_min    );
  test_comment("image max  = " << im_max    );
  test_comment("image mean = " << im_mean   );
  test_comment("image mean.r = " << (int)im_mean.r());
  test_comment("image mean.g = " << (int)im_mean.g());
  test_comment("image mean.b = " << (int)im_mean.b());
  test_comment("image mean.a = " << (int)im_mean.a());
  test_expect( im_min  == "#000" );
  test_expect( im_max  == "#fff" );
  test_expect( im_mean == "#7f7f7f" );
  test_expect( im_mean.r() >= 126 && im_mean.r() <= 128 );
  test_expect( im_mean.g() >= 126 && im_mean.g() <= 128 );
  test_expect( im_mean.b() >= 126 && im_mean.b() <= 128 );
  test_expect( im_mean.a() == 255 || im_mean.a() == 0   );
  {
    unsigned long n=0,r=0,g=0,b=0,a=0, sum;
    for(int i=0; i<im.width(); i++) {
      for(int j=0; j<im.height(); j++) {
        color_t col = im.color_at(i,j);
        n ++;
        r += col.r();
        g += col.g();
        b += col.b();
        a += col.a();
      }
    }
    sum = r+g+b;
    test_comment("num px= " << n );
    test_comment("acc px = " << sum );
    test_comment("mean R = " << (r/=n) );
    test_comment("mean G = " << (g/=n) );
    test_comment("mean B = " << (b/=n) );
    test_comment("mean A = " << (a/=n) );
    test_expect( (int)im_mean.r() >= (int)r-1 && (int)im_mean.r() <= (int)r+1 );
    test_expect( (int)im_mean.g() >= (int)g-1 && (int)im_mean.g() <= (int)g+1 );
    test_expect( (int)im_mean.b() >= (int)b-1 && (int)im_mean.b() <= (int)b+1 );
    test_expect( (int)im_mean.a() >= (int)a-1 && (int)im_mean.a() <= (int)a+1 );
  }
  // Split channels
  test_expect_noexcept(
    im.fill(color_t::black)
      .draw(point_t(2,0), color_t::red).draw(point_t(2,1), color_t::red).draw(point_t(2,2), color_t::red)
      .draw(point_t(1,0), color_t::green).draw(point_t(1,1), color_t::green)
      .draw(point_t(0,0), color_t::blue)
  );
  {
    cvimage::mat_vector_t channels = im.split_channels();
    test_expect( cvimage(channels[0]).num_nonzero_pixels() == 3 ); // R
    test_expect( cvimage(channels[1]).num_nonzero_pixels() == 2 ); // G
    test_expect( cvimage(channels[2]).num_nonzero_pixels() == 1 ); // B
  }
  test_expect( im.bounding_rect() == rect_t(0,0, 2,2) );
}
 
void test_modification()
{
  cvimage im(100,100, cvimage::csp_rgb);
  test_expect_noexcept(im.draw(rect_t(9,9,20,20)));
  test_expect_noexcept( im.crop(rect_t(10,10,19,19)) );
  test_expect(im.num_nonzero_pixels() == 0);
  test_comment("image size now:" << im.size());
  test_expect(im.width() == 10 && im.height() == 10);
 
  test_expect_noexcept( im.extend(1, cvimage::em_color, color_t::white) );
  test_expect(im.width() == 12 && im.height() == 12);
  test_expect(im.num_nonzero_pixels() == 4*11);
 
  test_expect_noexcept( im = im.subimage(rect_t(0,0,4,4)) );
  test_expect(im.width() == 5 && im.height() == 5);
  test_expect(im.num_nonzero_pixels() == 2*5-1);
 
  test_expect_noexcept( im.erode() );
  test_expect(im.num_nonzero_pixels() <= 1);
 
  test_expect_noexcept( im.fill(color_t::black).draw(point_t(2,2), color_t::white) );
  test_expect(im.num_nonzero_pixels() == 1);
  test_expect_noexcept( im.dilate() );
  test_expect(im.num_nonzero_pixels() >2 && im.num_nonzero_pixels() <= 9);
  test_expect_noexcept( im.dilate(10) );
  test_expect(im.num_nonzero_pixels() == im.area() );
 
  // resize / resample
  test_expect_noexcept( im.resample_up() );
  test_expect(im.size() == point_t(10,10) );
  test_expect_noexcept( im.resample_down() );
  test_expect(im.size() == point_t(5,5) );
  test_expect_noexcept( im.resize(20,20) );
  test_expect(im.size() == point_t(20,20) );
  test_expect_noexcept( im.resize_scale(2) );
  test_expect(im.size() == point_t(40,40) );
  test_expect_noexcept( im.resize_stretch(20,10) );
  test_expect(im.size() == point_t(20,10) );
 
  // noexcept checks
  test_expect_noexcept( im.blur() );
  test_expect_noexcept( im.blur(1) );
  test_expect_noexcept( im.blur(1, cvimage::blur_box ) );
  test_expect_noexcept( im.blur(1, cvimage::blur_gauss ) );
  test_expect_noexcept( im.blur(1, cvimage::blur_median ) );
  test_expect_noexcept( im.gradient() );
  test_expect_noexcept( im.gradient(2) );
  test_expect_noexcept( im.laplacian() );
  test_expect_noexcept( im.laplacian(1) );
  test_expect_noexcept( im.scharr(1,0) );
  test_expect_noexcept( im.scharr(0,1) );
  test_expect_noexcept( im.sobel(1,1) );
  test_expect_noexcept( im.sobel(0,1) );
  test_expect_noexcept( im.sobel(1,0) );
  test_expect_noexcept( im.sobel(1,0,2) );
  test_expect_noexcept( im.level() );
  test_expect_noexcept( im.level(0.1,0.9) );
  test_expect_noexcept( im.normalize(0.1,0.9) );
  test_expect_noexcept( im.invert() );
}
 
void test_contours()
{
  cvimage im(128,128, cvimage::csp_gray);
  im.fill(circle_t(64,64,32), color_t::white);
  im.fill(circle_t(64,64,16), color_t::black);
  // only outmost contours
  {
    contour_list_t cnt = im.contours(true);
    if(test_expect_cond( cnt.size() == 1 )) {
      rect_t rct;
      test_expect_noexcept( rct = cnt.front().bounding_rect() );
      test_expect( rct.tl() >= point_t(31,31) && rct.br() <= point_t(97,97) );
    }
  }
  // inner and outer in one list
  {
    contour_list_t cnt = im.contours(false);
    if(test_expect_cond( cnt.size() == 2 )) {
      rect_t rct;
      test_expect_noexcept( rct = cnt[0].bounding_rect() );
      test_expect(
        (rct.tl() >= point_t(31,31) && rct.br() <= point_t(97,97)) ||
        (rct.tl() >= point_t(46,46) && rct.br() <= point_t(82,82))
      );
      test_expect_noexcept( rct = cnt[1].bounding_rect() );
      test_expect(
        (rct.tl() >= point_t(31,31) && rct.br() <= point_t(97,97)) ||
        (rct.tl() >= point_t(46,46) && rct.br() <= point_t(82,82))
      );
    }
  }
  // contour comp (inner, outer)
  {
    contour_ccomp_list_t cnt = im.contours_ccomp();
    if(test_expect_cond( cnt.inner.size() == 1 && cnt.outer.size() == 1 )) {
      rect_t rct;
      test_expect_noexcept( rct = cnt.inner.front().bounding_rect() );
      test_expect( rct.tl() >= point_t(46,46) && rct.br() <= point_t(82,82));
      test_expect_noexcept( rct = cnt.outer.front().bounding_rect() );
      test_expect(rct.tl() >= point_t(31,31) && rct.br() <= point_t(97,97));
    }
  }
  // contour tree
  {
    contour_nested_list_t cnt = im.contours_tree();
    if(test_expect_cond( !cnt.has_parent() && !cnt.empty() )) { // the whole image
      rect_t rct;
      test_expect_noexcept( rct = cnt.val().bounding_rect() );
      test_expect(rct.tl() == point_t(0,0) && rct.br() == point_t(im.width()-1,im.height()-1));
      if(test_expect_cond( cnt.has_children() && cnt.children().size() == 1 )) { // outer circle
        cnt = cnt.children().front();
        test_expect_noexcept( rct = cnt.val().bounding_rect() );
        test_expect(rct.tl() >= point_t(31,31) && rct.br() <= point_t(97,97));
        if(test_expect_cond( cnt.has_children() && cnt.children().size() == 1 )) { // inner circle
          cnt = cnt.children().front();
          test_expect_noexcept( rct = cnt.val().bounding_rect() );
          test_expect( rct.tl() >= point_t(46,46) && rct.br() <= point_t(82,82));
        }
      }
      test_comment(rct.tl() << "/" << rct.br() );
    }
  }
}
 
void test_detectors()
{
  cvimage imo(128,128, cvimage::csp_gray);
  test_expect_noexcept(
    imo.fill(rect_t(point_t(32,32), 64,64), color_t::white)
       .fill(rect_t(point_t(56,56), 16,16), color_t::black)
  );
  {
    cvimage im(imo);
    test_expect_noexcept( im.corners_harris() );
    test_expect( im.max() == color_t::white );
    test_expect( im.min() == color_t::black );
    test_expect_noexcept ( im.threshold(color_t(im.max())) );
    test_expect( im.num_nonzero_pixels() == 8 );
  }
  {
    cvimage im(imo);
    test_expect_noexcept( im.edges_canny().threshold() );
    test_expect( im.num_nonzero_pixels() >= (4*62+4*14) && im.num_nonzero_pixels() <= (4*64+4*16) );
  }
  {
    cvimage im(imo);
    test_expect_noexcept(im.fill(color_t::black).fill(circle_t(30,30,25), color_t::white));
    test_expect_noexcept(im.fill(circle_t(80,80,25), color_t::white).to_grayscale());
    cvimage::circle_list_t circles;
    test_expect_noexcept( circles = im.blur(5, cvimage::blur_gauss).hough_circles(30,20));
    if(test_expect_cond( circles.size() == 2 )) {
      test_expect( circles.front().center() >= point_t(29,29) && circles.front().center() <= point_t(31,31));
      test_expect( circles.front().r() <=27 && circles.front().r() >=22 );
      test_expect( circles.back().center() >= point_t(79,79) && circles.back().center() <= point_t(81,81));
      test_expect( circles.back().r() <=27 && circles.back().r() >=22 );
    }
  }
}
 
void test()
{
  test_load_save();
  test_color();
  test_draw();
  test_conv();
  test_getters();
  test_modification();
  test_contours();
  test_detectors();
}