1、肤色侦测法 肤色提取是基于人机互动方面常见的方法。因为肤色是人体的一大特征,它可以迅速从复杂的背景下分离出自己的特征区域。一下介绍两种常见的肤色提取: (1)HSV空间的肤色提取 HSV色彩空间是一个圆锥形的模型,具体如右图所示: 色相(H)是色彩的基本属性,就是平常说的颜色名称,例如红色、黄色等, 依照右图的标准色轮上的位置,取360度得数值。(也有0~100%的方法确定) 饱和度(S)是色彩的纯度,越高色彩越纯,低则变灰。取值为0~100%。明度(V)也叫亮度,取值0~100。 根据肤色在HSV三个分量上的值,就可以简单的侦测出一张图像上肤色的部分。一下是肤色侦测函数的源代码: [c-sharp] view plaincopy void skinDetectionHSV(IplImage* pImage,int lower,int upper,IplImage* process) { IplImage* pImageHSV = NULL; IplImage* pImageH = NULL; IplImage* pImageS = NULL; IplImage* pImageProcessed = NULL; IplImage* tmpH = NULL; IplImage* tmpS = NULL; static IplImage* pyrImage = NULL; CvSize imgSize; imgSize.height = pImage->height; imgSize.width = pImage->width ; //create you want to use image and give them memory allocation pImageHSV = cvCreateImage(imgSize,IPL_DEPTH_8U,3); pImageH = cvCreateImage(imgSize,IPL_DEPTH_8U,1); pImageS = cvCreateImage(imgSize,IPL_DEPTH_8U,1); tmpS = cvCreateImage(imgSize,IPL_DEPTH_8U,1); tmpH = cvCreateImage(imgSize,IPL_DEPTH_8U,1); pImageProcessed = cvCreateImage(imgSize,IPL_DEPTH_8U,1); pyrImage = cvCreateImage(cvSize(pImage->width/2,pImage->height/2),IPL_DEPTH_8U,1); //convert RGB image to HSV image cvCvtColor(pImage,pImageHSV,CV_BGR2HSV); //Then split HSV to three single channel images cvCvtPixToPlane(pImageHSV,pImageH,pImageS,NULL,NULL); //The skin scalar range in H and S, Do they AND algorithm cvInRangeS(pImageH,cvScalar(0.0,0.0,0,0),cvScalar(lower,0.0,0,0),tmpH); cvInRangeS(pImageS,cvScalar(26,0.0,0,0),cvScalar(upper,0.0,0,0),tmpS); cvAnd(tmpH,tmpS,pImageProcessed,0); // //cvPyrDown(pImageProcessed,pyrImage,CV_GAUSSIAN_5x5); //cvPyrUp(pyrImage,pImageProcessed,CV_GAUSSIAN_5x5); //Erode and dilate cvErode(pImageProcessed,pImageProcessed,0,2); cvDilate(pImageProcessed,pImageProcessed,0,1); cvCopy(pImageProcessed,process,0); //do clean cvReleaseImage(&pyrImage); cvReleaseImage(&pImageHSV); cvReleaseImage(&pImageH); cvReleaseImage(&pImageS); cvReleaseImage(&pyrImage); cvReleaseImage(&tmpH); cvReleaseImage(&tmpS); cvReleaseImage(&pImageProcessed); } (2)YCrCb空间的肤色提取 YCrCb也是一种颜色空间,也可以说是YUV的颜色空间。Y是亮度的分量,而肤色侦测是对亮度比较敏感的,由摄像头拍摄的RGB图像转化为YCrCb空间的话可以去除亮度对肤色侦测的影响。下面给出基于YCrCb肤色侦测函数的源代码: [c-sharp] view plaincopy void skinDetectionYCrCb(IplImage* imageRGB,int lower,int upper,IplImage* imgProcessed) { assert(imageRGB->nChannels==3); IplImage* imageYCrCb = NULL; IplImage* imageCb = NULL; imageYCrCb = cvCreateImage(cvGetSize(imageRGB),8,3); imageCb = cvCreateImage(cvGetSize(imageRGB),8,1); cvCvtColor(imageRGB,imageYCrCb,CV_BGR2YCrCb); cvSplit(imageYCrCb,0,0,imageCb,0);//Cb for (int h=0;hheight;h++) { for (int w=0;wwidth;w++) { unsigned char* p =(unsigned char*)(imageCb->imageData+h*imageCb->widthStep+w); if (*p<=upper&&*p>=lower) { *p=255; } else { *p=0; } } } cvCopy(imageCb,imgProcessed,NULL); } 2、基于混合高斯模型去除背景法 高斯模型去除背景法也是背景去除的一种常用的方法,经常会用到视频图像侦测中。这种方法对于动态的视频图像特征侦测比较适合,因为模型中是前景和背景分离开来的。分离前景和背景的基准是判断像素点变化率,会把变化慢的学习为背景,变化快的视为前景。 [c-sharp] view plaincopy // #include "stdafx.h" #include "cv.h" #include "highgui.h" #include "cxtypes.h" #include "cvaux.h" # include using namespace std; int _tmain(int argc, _TCHAR* argv[]) { //IplImage* pFirstFrame = NULL; IplImage* pFrame = NULL; IplImage* pFrImg = NULL; IplImage* pBkImg = NULL; IplImage* FirstImg = NULL; static IplImage* pyrImg =NULL; CvCapture* pCapture = NULL; int nFrmNum = 0; int first = 0,next = 0; int thresh = 0; cvNamedWindow("video",0); //cvNamedWindow("background",0); cvNamedWindow("foreground",0); cvResizeWindow("video",400,400); cvResizeWindow("foreground",400,400); //cvCreateTrackbar("thresh","foreground",&thresh,255,NULL); //cvMoveWindow("background",360,0); //cvMoveWindow("foregtound",0,0); if(!(pCapture = cvCaptureFromCAM(1))) { printf("Could not initialize camera , please check it !"); return -1; } CvGaussBGModel* bg_model = NULL; while(pFrame = cvQueryFrame(pCapture)) { nFrmNum++; if(nFrmNum == 1) { pBkImg = cvCreateImage(cvGetSize(pFrame),IPL_DEPTH_8U,3); pFrImg = cvCreateImage(cvGetSize(pFrame),IPL_DEPTH_8U,1); FirstImg = cvCreateImage(cvGetSize(pFrame),IPL_DEPTH_8U,1); pyrImg = cvCreateImage(cvSize(pFrame->width/2,pFrame->height/2),IPL_DEPTH_8U,1); CvGaussBGStatModelParams params; params.win_size = 2000; //Learning rate = 1/win_size; params.bg_threshold = 0.7; //Threshold sum of weights for background test params.weight_init = 0.05; params.variance_init = 30; params.minArea = 15.f; params.n_gauss = 5; //= K =Number of gaussian in mixture params.std_threshold = 2.5; //cvCopy(pFrame,pFirstFrame,0); bg_model = (CvGaussBGModel*)cvCreateGaussianBGModel(pFrame,¶ms); } else { int regioncount = 0; int totalNum = pFrImg->width *pFrImg->height ; cvSmooth(pFrame,pFrame,CV_GAUSSIAN,3,0,0,0); cvUpdateBGStatModel(pFrame,(CvBGStatModel*)bg_model,-0.00001); cvCopy(bg_model->foreground ,pFrImg,0); cvCopy(bg_model->background ,pBkImg,0); //cvShowImage("background",pBkImg); //cvSmooth(pFrImg,pFrImg,CV_GAUSSIAN,3,0,0,0); //cvPyrDown(pFrImg,pyrImg,CV_GAUSSIAN_5x5); //cvPyrUp(pyrImg,pFrImg,CV_GAUSSIAN_5x5); //cvSmooth(pFrImg,pFrImg,CV_GAUSSIAN,3,0,0,0); cvErode(pFrImg,pFrImg,0,1); cvDilate(pFrImg,pFrImg,0,3); //pBkImg->origin = 1; //pFrImg->origin = 1; cvShowImage("video",pFrame); cvShowImage("foreground",pFrImg); //cvReleaseBGStatModel((CvBGStatModel**)&bg_model); //bg_model = (CvGaussBGModel*)cvCreateGaussianBGModel(pFrame,0); /* //catch target frame if(nFrmNum>10 &&(double)cvSumImage(pFrImg)>0.3 * totalNum) { first = cvSumImage(FirstImg); next = cvSumImage(pFrImg); printf("Next number is :%d /n",next); cvCopy(pFrImg,FirstImg,0); } cvShowImage("foreground",pFrImg); cvCopy(pFrImg,FirstImg,0); */ if(cvWaitKey(2)== 27) { break; } } } cvReleaseBGStatModel((CvBGStatModel**)&bg_model); cvDestroyAllWindows(); cvReleaseImage(&pFrImg); cvReleaseImage(&FirstImg); cvReleaseImage(&pFrame); cvReleaseImage(&pBkImg); cvReleaseCapture(&pCapture); return 0; } 3、背景相减背景去除方法 所谓的背景相减,是指把摄像头捕捉的图像第一帧作为背景,以后的每一帧都减去背景帧,这样减去之后剩下的就是多出来的特征物体(要侦测的物体)的部分。但是相减的部分也会对特征物体的灰阶值产生影响,一般是设定相关阈值要进行判断。以下是代码部分: [c-sharp] view plaincopy int _tmain(int argc, _TCHAR* argv[]) { int thresh_low = 30; IplImage* pImgFrame = NULL; IplImage* pImgProcessed = NULL; IplImage* pImgBackground = NULL; IplImage* pyrImage = NULL; CvMat* pMatFrame = NULL; CvMat* pMatProcessed = NULL; CvMat* pMatBackground = NULL; CvCapture* pCapture = NULL; cvNamedWindow("video", 0); cvNamedWindow("background",0); cvNamedWindow("processed",0); //Create trackbar cvCreateTrackbar("Low","processed",&thresh_low,255,NULL); cvResizeWindow("video",400,400); cvResizeWindow("background",400,400); cvResizeWindow("processed",400,400); cvMoveWindow("video", 0, 0); cvMoveWindow("background", 400, 0); cvMoveWindow("processed", 800, 0); if( !(pCapture = cvCaptureFromCAM(1))) { fprintf(stderr, "Can not open camera./n"); return -2; } //first frame pImgFrame = cvQueryFrame( pCapture ); pImgBackground = cvCreateImage(cvSize(pImgFrame->width, pImgFrame->height), IPL_DEPTH_8U,1); pImgProcessed = cvCreateImage(cvSize(pImgFrame->width, pImgFrame->height), IPL_DEPTH_8U,1); pyrImage = cvCreateImage(cvSize(pImgFrame->width/2, pImgFrame->height/2), IPL_DEPTH_8U,1); pMatBackground = cvCreateMat(pImgFrame->height, pImgFrame->width, CV_32FC1); pMatProcessed = cvCreateMat(pImgFrame->height, pImgFrame->width, CV_32FC1); pMatFrame = cvCreateMat(pImgFrame->height, pImgFrame->width, CV_32FC1); cvSmooth(pImgFrame, pImgFrame, CV_GAUSSIAN, 3, 0, 0); cvCvtColor(pImgFrame, pImgBackground, CV_BGR2GRAY); cvCvtColor(pImgFrame, pImgProcessed, CV_BGR2GRAY); cvConvert(pImgProcessed, pMatFrame); cvConvert(pImgProcessed, pMatProcessed); cvConvert(pImgProcessed, pMatBackground); cvSmooth(pMatBackground, pMatBackground, CV_GAUSSIAN, 3, 0, 0); while(pImgFrame = cvQueryFrame( pCapture )) { cvShowImage("video", pImgFrame); cvSmooth(pImgFrame, pImgFrame, CV_GAUSSIAN, 3, 0, 0); cvCvtColor(pImgFrame, pImgProcessed, CV_BGR2GRAY); cvConvert(pImgProcessed, pMatFrame); cvSmooth(pMatFrame, pMatFrame, CV_GAUSSIAN, 3, 0, 0); cvAbsDiff(pMatFrame, pMatBackground, pMatProcessed); //cvConvert(pMatProcessed,pImgProcessed); //cvThresholdBidirection(pImgProcessed,thresh_low); cvThreshold(pMatProcessed, pImgProcessed, 30, 255.0, CV_THRESH_BINARY); cvPyrDown(pImgProcessed,pyrImage,CV_GAUSSIAN_5x5); cvPyrUp(pyrImage,pImgProcessed,CV_GAUSSIAN_5x5); //Erode and dilate cvErode(pImgProcessed, pImgProcessed, 0, 1); cvDilate(pImgProcessed, pImgProcessed, 0, 1); //background update cvRunningAvg(pMatFrame, pMatBackground, 0.0003, 0); cvConvert(pMatBackground, pImgBackground); cvShowImage("background", pImgBackground); cvShowImage("processed", pImgProcessed); //cvZero(pImgProcessed); if( cvWaitKey(10) == 27 ) { break; } } cvDestroyWindow("video"); cvDestroyWindow("background"); cvDestroyWindow("processed"); cvReleaseImage(&pImgProcessed); cvReleaseImage(&pImgBackground); cvReleaseMat(&pMatFrame); cvReleaseMat(&pMatProcessed); cvReleaseMat(&pMatBackground); cvReleaseCapture(&pCapture); return 0; }
转载于:https://www.cnblogs.com/ligun123/archive/2012/04/19/2457433.html