opencv_test.rar_CENSURE OPENCV_CenSurE

#include "CenSurE.h" double gaussian(double sigma, double x) { return (1/(sigma*sqrt(2*PI)))*exp(-x*x/(sigma*sigma)); } vector <SDescriptor> GetFeature(IplImage *original_image) { double tt; //0. Image Load int i, j; int x, y; int width, height; width = original_image->width; height = original_image->height; IplImage *gray = cvCreateImage(cvGetSize(original_image), 8, 1); IplImage *i_image = cvCreateImage(cvSize(original_image->width +1, original_image->height + 1), IPL_DEPTH_32S, 1); cvCvtColor(original_image, gray, CV_BGR2GRAY); cvIntegral(gray, i_image); //1. DoB filter int min_n = 1; int max_n = 7; int n = min_n;//min_n ~ max_n double max_value = 0;//가장 큰 DoB값, 나중에 normalize하기위해? double out_value = 0;//DoB구할 때 바깥 double in_value = 0;//DoB구할 때 안쪽 double I_n = 1; double O_n = 0; double DoB = 0; SDoBInfo **m_DoB_info; m_DoB_info = (SDoBInfo **)malloc(height*sizeof(SDoBInfo *)); for(i = 0; i < height; i++) { m_DoB_info[i] = (SDoBInfo *)malloc(width*sizeof(SDoBInfo)); } //printf("%f\n", max_value); for(int n = min_n; n <= max_n; n++) { I_n = I_n*(2*n + 1)*(2*n + 1)/((2*n + 3)*(2*n + 3)); O_n = I_n*(2*n + 1)*(2*n + 1)/((4*n + 1)*(4*n + 1)); //printf("%d I = %f O = %f\n", n, I_n, O_n); for(int y = 2*max_n + 1; y < height - 2*max_n; y++) { for(int x = 2*max_n + 1; x < width - 2*max_n; x++) { out_value = cvGetReal2D(i_image, y + 2*n, x + 2*n) - cvGetReal2D(i_image, y - 2*n - 1, x + 2*n) - cvGetReal2D(i_image, y + 2*n, x - 2*n - 1) + cvGetReal2D(i_image, y - 2*n - 1, x - 2*n - 1); in_value = cvGetReal2D(i_image, y + n, x + n) - cvGetReal2D(i_image, y - n - 1, x + n) - cvGetReal2D(i_image, y + n, x - n - 1) + cvGetReal2D(i_image, y - n - 1, x - n - 1); DoB = O_n*out_value - I_n*in_value; m_DoB_info[y][x].DoB[n-min_n] = DoB; if(DoB > max_value) { max_value = DoB; } } } } tt = (double)cvGetTickCount(); //2. Non Max suppression double current_value; int current_n, current_x, current_y; int max_ok = 1; vector <SDescriptor> m_descriptors; SDescriptor descriptor;//임시 int gap = 1; for(int n = min_n+gap; n <= max_n-gap; n++) { for(int y = 2*max_n + 1 + gap; y < height - 2*max_n - gap; y++) { for(int x = 2*max_n + 1 + gap; x < width - 2*max_n - gap; x++) { DoB = m_DoB_info[y][x].DoB[n-1]; if(DoB > max_value/5.0) { max_ok = 1; for(current_n = n-1; current_n <= n+1; current_n++) { for(current_y = y-1; current_y <= y+1; current_y++) { for(current_x = x-1; current_x <= x+1; current_x++) { current_value = m_DoB_info[current_y][current_x].DoB[current_n-1]; if(current_value > DoB) { max_ok = 0; } } } } if(max_ok == 1) { //printf("%d %d\n", x, y); cvCircle(original_image, cvPoint(x, y), 3, cvScalar(0, 0, 255),1); descriptor.x = x; descriptor.y = y; m_descriptors.push_back(descriptor); } } } } } tt = (double)cvGetTickCount() - tt; printf( "basic time = %gms\n", tt/(cvGetTickFrequency()*1000.)); //printf("%d", m_descriptors.size()); double descriptor_dx[24][24]; double descriptor_dy[24][24]; double descriptor_adx[24][24]; double descriptor_ady[24][24]; int region_half_width = 12; int region_width = 24; int start_x, end_x; int start_y, end_y; double dx; double dy; int sub_region_width = 9; int sub_region_half_width = 4; int sub_region_x, sub_region_y; double dist; double weight; double sum_dx; double sum_dy; double sum_adx; double sum_ady; int start_sub_x, end_sub_x; int start_sub_y, end_sub_y; int descriptor_index; double sum_descriptor; for(i=0; i < m_descriptors.size(); i++) { //양쪽 2s씩 padding if( (m_descriptors[i].x >= 2*max_n + 1 + region_half_width) && (m_descriptors[i].x < width - 2*max_n - region_half_width) && (m_descriptors[i].y >= 2*max_n + 1 + region_half_width) && (m_descriptors[i].y < height - 2*max_n - region_half_width)) { start_x = m_descriptors[i].x - region_half_width; end_x = m_descriptors[i].x + region_half_width; start_y = m_descriptors[i].y - region_half_width; end_y = m_descriptors[i].y + region_half_width; for(y = start_y; y < end_y; y++) { for(x = start_x; x < end_x; x++) { dx = cvGetReal2D(gray, y, x+1) - cvGetReal2D(gray, y, x); dy = cvGetReal2D(gray, y+1, x) - cvGetReal2D(gray, y, x); descriptor_dx[y-start_y][x-start_x] = dx; descriptor_dy[y-start_y][x-start_x] = dy; descriptor_adx[y-start_y][x-start_x] = fabs(dx); descriptor_ady[y-start_y][x-start_x] = fabs(dy); } } descriptor_index = 0; for(sub_region_y = start_y + 5; sub_region_y < end_y; sub_region_y+=5) { for(sub_region_x = start_x + 5; sub_region_x < end_x; sub_region_x+=5) { start_sub_x = sub_region_x - sub_region_half_width; end_sub_x = sub_region_x + sub_region_half_width; start_sub_y = sub_region_y - sub_region_half_width; end_sub_y = sub_region_y + sub_region_half_width; sum_dx = 0; sum_dy = 0; sum_adx = 0; sum_ady = 0; for(y = start_sub_y; y <= end_sub_y; y++) { for(x = start_sub_x; x <= start_sub_x; x++) { dist = sqrt( (double)((x-sub_region_x)*(x-sub_region_x) + (y-sub_region_y)*(y-sub_region_y))); weight = gaussian(2.5, dist); //weight = 1; sum_dx += weight*descriptor_dx[y-start_sub_y][x-start_sub_x]; sum_dy += weight*descriptor_dy[y-start_sub_y][x-start_sub_x]; sum_adx += weight*descriptor_adx[y-start_sub_y][x-start_sub_x]; sum_ady += weight*descriptor_ady[y-start_sub_y][x-start_sub_x]; } } dist = sqrt( (double)((sub_region_x - m_descriptors[i].x)*(sub_region_x - m_descriptors[i].x) + (sub_region_y - m_descriptors[i].y)*(sub_region_y - m_descriptors[i].y))); weight = gaussian(1.3, dist); //weight = 1; m_descriptors[i].v[descriptor_index] = weight*sum_dx; m_descriptors[i].v[descriptor_index+1] = weight*sum_dy; m_descriptors[i].v[descriptor_index+2] = weight*sum_ady; m_descriptors[i].v[descriptor_index+3] = weight*sum_ady; descriptor_index += 4; } } sum_descriptor = 0; for(j=0; j<64;j++) { sum_descriptor += fabs(m_descriptors[i].v[j]); } for(j=0; j<64;j++) { m_descriptors[i].v[j] = m_descriptors[i].v[j]/sum_descriptor; } } } for(i = 0; i < height; i++) { free(m_DoB_info[i]); } free(m_DoB_info); cvReleaseImage(&gray); cvReleaseImage(&i_image); //vector <SDescriptor> m_descriptors; return m_descriptors; }