用于分割显着彩色图像和确定掌握的代码_C++_代码_下载

//g++ -std=c++11 *.cpp `pkg-config --libs --cflags opencv` -o main #include <opencv2/core.hpp> #include <opencv2/highgui/highgui.hpp> #include <opencv2/imgproc/imgproc.hpp> #include <opencv2/ximgproc/slic.hpp> #include <opencv2/core/cvstd.hpp> #include <opencv2/core/utility.hpp> #include <opencv2/ximgproc.hpp> #include <opencv2/saliency.hpp> #include <iostream> #include <vector> #include <map> #include <string> #include <stdio.h> #include <math.h> #include <float.h> #include "grasp.h" #include "line.h" using namespace std; using namespace cv; using namespace saliency; void getSuperPixels(const Mat &inputImg, Mat &outputImg, Mat &labels, int pixelSize); void getSaliencyMap(Mat &inputImg, Mat &outputImg); void binarizeSaliencyMap(const Mat &salMap, const Mat &pxlLabels, Mat &outputImg, double Thold); void dilateBinary(const Mat &binaryImg, Mat &dilation, int pixelSize); void cleanUpBinary(Mat &binaryImg); int main( int argc, char* argv[] ) { Mat image; //variable holds main image Mat BlurImage; //Image holds alterations Mat gCutImg; //super pixel image Mat spxlImg; //Saliency image Mat spxlLabel; Mat salImg; Mat dilatedImg; Mat fgd; Mat bgd; Mat binMask; String file = argv[1];//grab image from command line /* parameters of the algorithm */ int SPXLSIZE = 15; //average super pixel size double Tvalue = 0.25; //threshold valie //intialize the primary image image = imread(file , IMREAD_COLOR); /* get information on the image, height and width */ int imgH = image.rows; int imgW = image.cols; /* initialize the binarized image as all black */ Mat spxlSal = Mat::zeros(imgH,imgW,CV_8UC1); /* Blur the image with 3x3 Gaussian */ GaussianBlur(image,BlurImage,Size(5,5),0,0); /* Construct superpixel and generate mask*/ getSuperPixels(image, spxlImg, spxlLabel, SPXLSIZE); /* Compute the saliency map for the image using OpenCV */ getSaliencyMap(image, salImg); /* perfomr binarization on saliency map */ binarizeSaliencyMap(salImg,spxlLabel,spxlSal,Tvalue); /* Perform dilation on image */ dilateBinary(spxlSal,dilatedImg,SPXLSIZE); /* Find contours of the image */ vector<vector<Point> > contours; vector<Vec4i> hierarchy; findContours(spxlSal,contours,hierarchy,RETR_TREE,CHAIN_APPROX_SIMPLE,Point(0,0)); /* Find contours with largest area */ double area = 0; vector<Point> *tracker; vector<vector<Point>>::iterator cntItr = contours.begin(); for(;cntItr != contours.end();cntItr++){ double tmpArea = contourArea(*cntItr); if(tmpArea > area) { tracker = &(*cntItr); area = tmpArea; } } /* Find the bounding box of the largest contour */ Rect bnds = boundingRect(*tracker); /* pad rectangle */ int padSize=10; if(bnds.x-padSize >= 0) { bnds.x = bnds.x - 10; if(bnds.width + 2*padSize < imgW) { bnds.width = bnds.width + 2*padSize; } } if(bnds.y-padSize >= 0) { bnds.y = bnds.y - 10; if(bnds.height + 2*padSize < imgH) { bnds.height = bnds.height + 2*padSize; } } /* Use the above rectangle to grab cut the object */ grabCut(BlurImage,gCutImg,bnds,bgd,fgd,3,GC_INIT_WITH_RECT); binMask.create(gCutImg.size(),CV_8UC1); binMask = gCutImg & 1; /* Get contours and keep only the largest one */ cleanUpBinary(binMask); /* Make Grasp Object */ Grasp grasp(binMask); grasp.displayPolygon(); grasp.discretizePolygon(0.1); grasp.displayDiscretizedPolygon(); grasp.generateGrasp(); /* place super pixel mask on image*/ Mat spxlMod = image.clone(); binMask = binMask*255; cvtColor(binMask,binMask,COLOR_GRAY2BGR ); int i,j; Vec3b* p; Vec3b* q; for( i = 0; i < imgH; ++i) { p = binMask.ptr<Vec3b>(i); for ( j = 0; j < imgW; ++j) { if(p[j][0] == 255 || p[j][1] == 255 || p[j][2] == 255){ p[j][0] == 0; p[j][1] == 0; p[j][2] == 255; } } } for( i = 0; i < imgH; ++i) { p = binMask.ptr<Vec3b>(i); q = spxlMod.ptr<Vec3b>(i); for ( j = 0; j < imgW; ++j) { if(p[j][0] == 255 || p[j][1] == 255 || p[j][2] == 255){ q[j][0] = 0; q[j][1] = 0; q[j][2] = 255; } } } namedWindow("two", WINDOW_NORMAL); imshow("two",binMask); namedWindow("three", WINDOW_NORMAL); imshow("three",salImg); namedWindow("five", WINDOW_NORMAL); imshow("five",spxlSal); namedWindow("four", WINDOW_NORMAL); imshow("four",spxlMod); waitKey(0); } /*****************************************/ /* Functions for program below */ /*****************************************/ /* Following finds the superpixels and its labels in image inputImg and and places them in outputImg and labels. the average superpixel size will be set to pixelSize*/ void getSuperPixels(const Mat & inputImg, Mat & outputImg, Mat & labels, int pixelSize) { /* Blur the image with 3x3 Gaussian */ GaussianBlur(inputImg,outputImg,Size(5,5),0,0); /* Convert to LAB color space */ cvtColor(outputImg,outputImg, 44); /* Construct superpixel and generate mask*/ Ptr<ximgproc::SuperpixelSLIC> ptr = ximgproc::createSuperpixelSLIC(outputImg,100,pixelSize,75.0f); ptr->iterate(10); ptr->enforceLabelConnectivity(10); //reduces small fragments ptr->getLabelContourMask(outputImg,true); //returns contour mask ptr->getLabels(labels); //assigns labels } void getSaliencyMap(Mat &inputImg, Mat &outputImg) { /* Convert to LAB color space */ cvtColor(inputImg,outputImg, 44); Mat average = outputImg.clone(); /* Blur the image with 3x3 Gaussian */ GaussianBlur(outputImg,outputImg,Size(5,5),0,0); Ptr<Saliency> fine = StaticSaliencyFineGrained::create(); if(!fine->computeSaliency(outputImg,outputImg)){ cout << "error computing sliency" << endl; } normalize(outputImg,outputImg,0,255,NORM_MINMAX,0,noArray()); /* Find average value */ double li = 0; double a = 0; double b = 0; int numpxls = outputImg.rows*outputImg.cols; int i,j; Vec<unsigned char, 3>* p; for( i = 0; i < outputImg.rows; ++i) { p = outputImg.ptr<Vec<unsigned char, 3>>(i); for ( j = 0; j < outputImg.cols; ++j) { li += p[j][0]; a += p[j][1]; b += p[j][2]; } } li = li/(numpxls); a = a/(numpxls); b = b/(numpxls); // cout << li <<"\n"<<a <<"\n" <<b <<endl; Vec3b *q; Mat tmpimag = Mat::zeros(outputImg.rows,outputImg.cols,CV_8UC1); for( i = 0; i < average.rows; ++i) { q = outputImg.ptr<Vec3b>(i); for ( j = 0; j < average.cols; ++j) { q[j][0] = li-q[j][0]; q[j][1] = a-q[j][1]; q[j][2] = b-q[j][2]; } } outputImg = abs(outputImg); } void binarizeSaliencyMap(const Mat &salMap, const Mat &pxlLabels, Mat &outputImg, double Thold) { map<int,vector<int>> values; map<int,vector<Point>> locations; int imgH = salMap.rows; int imgW = salMap.cols; double min, max; minMaxLoc(salMap, &min, &max); double minPxl, spxlNum; minMaxLoc(pxlLabels, &minPxl, &spxlNum); const int *imgPtr; const unsigned char *salPtr; for(int i = 0; i < imgH; i++){ imgPtr = pxlLabels.ptr<const int>(i); for(int j = 0; j < imgW; j++){ //int curLbl = pxlLabels.at<int>(i, j); int curLbl = imgPtr[j]; //if the label is not in the map, add it. if(values.find(curLbl) == values.end()) { vector<int> tempVals; vector<Point> tempPtns; values.insert(pair<int,vector<int>>(curLbl,tempVals)); locations.insert(pair<int,vector<Point>>(curLbl,tempPtns)); } Point ptn; ptn.x = j; ptn.y = i; salPtr = salMap.ptr<const unsigned char>(i); int val = salPtr[j]; //int val = salMap.at<unsigned char>(i,j); values[curLbl].push_back(val); locations[curLbl].push_back(ptn); } } //check which superpixels pass the threshold map<int,vector<int>