// Copyright (C) 2016 Davis E. King (davis@dlib.net) // License: Boost Software License See LICENSE.txt for the full license. #ifndef DLIB_RANDOM_cOLOR_TRANSFORM_Hh_ #define DLIB_RANDOM_cOLOR_TRANSFORM_Hh_ #include "random_color_transform_abstract.h" #include "../image_processing/generic_image.h" #include "../pixel.h" #include "../rand.h" namespace dlib { // ---------------------------------------------------------------------------------------- class random_color_transform { public: random_color_transform ( dlib::rand& rnd, const double gamma_magnitude = 0.5, const double color_magnitude = 0.2 ) { // pick a random gamma correction factor. double gamma = std::max(0.0, 1 + gamma_magnitude*(rnd.get_random_double()-0.5)); // pick a random color balancing scheme. double red_scale = 1-rnd.get_random_double()*color_magnitude; double green_scale = 1-rnd.get_random_double()*color_magnitude; double blue_scale = 1-rnd.get_random_double()*color_magnitude; const double m = 255*std::max(std::max(red_scale,green_scale),blue_scale); red_scale /= m; green_scale /= m; blue_scale /= m; // Now compute a lookup table for all the color channels. The table tells us // what the transform does. table.resize(256*3); unsigned long i = 0; for (int k = 0; k < 256; ++k) { double v = 255*std::pow(k*red_scale, gamma); table[i++] = (unsigned char)(v + 0.5); } for (int k = 0; k < 256; ++k) { double v = 255*std::pow(k*green_scale, gamma); table[i++] = (unsigned char)(v + 0.5); } for (int k = 0; k < 256; ++k) { double v = 255*std::pow(k*blue_scale, gamma); table[i++] = (unsigned char)(v + 0.5); } } rgb_pixel operator()(rgb_pixel p) const { p.red = table[(unsigned int)p.red]; p.green = table[(unsigned int)p.green+256]; p.blue = table[(unsigned int)p.blue+512]; return p; } private: std::vector<unsigned char> table; }; // ---------------------------------------------------------------------------------------- template <typename image_type> void disturb_colors ( image_type& img_, dlib::rand& rnd, const double gamma_magnitude = 0.5, const double color_magnitude = 0.2 ) { image_view<image_type> img(img_); random_color_transform tform(rnd, gamma_magnitude, color_magnitude); for (long r = 0; r < img.nr(); ++r) { for (long c = 0; c < img.nc(); ++c) { rgb_pixel temp; assign_pixel(temp, img[r][c]); temp = tform(temp); assign_pixel(img[r][c], temp); } } } // ---------------------------------------------------------------------------------------- template <typename image_type> void apply_random_color_offset ( image_type& img_, dlib::rand& rnd ) { // Make a random color offset. This tform matrix came from looking at the // covariance matrix of RGB values in a bunch of images. In particular, if you // multiply Gaussian random vectors by tform it will result in vectors with the // same covariance matrix as the original RGB data. Also, this color transform is // what is suggested by the paper: // Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet // classification with deep convolutional neural networks." Advances in neural // information processing systems. 2012. // Except that we used the square root of the eigenvalues (which I'm pretty sure is // what the authors intended). matrix<double,3,3> tform; tform = -66.379, 25.094, 6.79698, -68.0492, -0.302309, -13.9539, -68.4907, -24.0199, 7.27653; matrix<double,3,1> v; v = rnd.get_random_gaussian(),rnd.get_random_gaussian(),rnd.get_random_gaussian(); v = round(tform*0.1*v); const int roffset = v(0); const int goffset = v(1); const int boffset = v(2); // Make up lookup tables that apply the color mapping so we don't have to put a // bunch of complicated conditional branches in the loop below. unsigned char rtable[256]; unsigned char gtable[256]; unsigned char btable[256]; for (int i = 0; i < 256; ++i) { rtable[i] = put_in_range(0, 255, i+roffset); gtable[i] = put_in_range(0, 255, i+goffset); btable[i] = put_in_range(0, 255, i+boffset); } // now transform the image. image_view<image_type> img(img_); for (long r = 0; r < img.nr(); ++r) { for (long c = 0; c < img.nc(); ++c) { rgb_pixel temp; assign_pixel(temp, img[r][c]); temp.red = rtable[temp.red]; temp.green = gtable[temp.green]; temp.blue = btable[temp.blue]; assign_pixel(img[r][c], temp); } } } // ---------------------------------------------------------------------------------------- } #endif // DLIB_RANDOM_cOLOR_TRANSFORM_Hh_