#ifndef STIM_CUDA_KMEANS_CUH #define STIM_CUDA_KMEANS_CUH #include //#include #include //#include #include #include /// This function use cvkmeans to cluster given textons /// @param testons is a multi-channel image /// @param K is the number of clusters stim::image kmeans(stim::image textons, unsigned int K){ unsigned int w = textons.width(); // get the width of picture unsigned int h = textons.height(); // get the height of picture unsigned int feature_n = textons.channels(); // get the spectrum of picture unsigned int N = w * h; // get the number of pixels float* sample1 = (float*) malloc(sizeof(float) * N * feature_n); //allocate the space for textons //reallocate a multi-channel texton image to a single-channel image for(unsigned int c = 0; c < feature_n; c++){ stim::image temp; temp = textons.channel(c); for(unsigned int j = 0; j < N; j++){ sample1[c + j * feature_n] = temp.data()[j]; } } cv::Mat sample2(N, feature_n, CV_32F, sample1); //copy image to cv::mat //(optional) show the test result //imshow("sample2", sample2); cv::TermCriteria criteria(CV_TERMCRIT_EPS+CV_TERMCRIT_ITER, 10, 0.1); // set stop-criteria for kmeans iteration cv::Mat labels(N, 1, CV_8U, cvScalarAll(0)); // allocate space for kmeans output cv::Mat centers; // allocate space for kmeans output unsigned int test_times = 2; // set the number of times of trying kmeans, it will return the best result cv::kmeans(sample2, K, labels, criteria, test_times, cv::KMEANS_PP_CENTERS, centers); // kmeans clustering //(optional) show the test result //imwrite( "data_output/labels_1D.bmp", labels); stim::image texture(w, h, 1, 1); // allocate space for texture for(unsigned int i = 0; i < N; i++){ // reshape the labels from iD array to image texture.data()[i] = labels.at(i); } //texture.save("data_output/kmeans_test0924_2.bmp"); //(optional) show the test result //stim::cpu2image(texture.data(), "data_output/kmeans_test.bmp", w, h, stim::cmBrewer); return texture; } #endif