终于到了有实际应用的功能了,有2张图片,里面各有一个人脸,我的目的是比较这两个人脸的相似度,这里用到了facedetect的功能,还有图像转换,图像剪切,以及直方图的比较。具体流程是:
1。分别用facedetect功能将两张图片中的人脸检测出来
2。将人脸部分的图片剪切出来,存到两张只有人脸的图片里。
3。将这两张人脸图片转换成单通道的图像
4。使用直方图比较这两张单通道的人脸图像,得出相似度。
这里对图的要求还是比较高的,光线和姿势不能有差别,脸的垂直或者左右角度偏差就会影响比较,但和两张图片的大小关系不大,本人觉得较适合于证件照的对比。
下面是代码,其中haarcascade_frontalface_alt.xml是opencv里facedetect例子用的样本。 比较的是srcImage和targetImage对应的文件.
还有下面是IplImage和Mat混用,纯当熟悉这两个类了。
#include "opencv/cv.hpp"
#include "opencv2/objdetect/objdetect.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include
#include
using namespace std;
using namespace cv;
String cascadeName = "D:\\OpenCV-2.4.2\\data\\haarcascades\\haarcascade_frontalface_alt.xml";
IplImage* cutImage(IplImage* src, CvRect rect) {
cvSetImageROI(src, rect);
IplImage* dst = cvCreateImage(cvSize(rect.width, rect.height),
src->depth,
src->nChannels);
cvCopy(src,dst,0);
cvResetImageROI(src);
return dst;
}
IplImage* detect( Mat& img, CascadeClassifier& cascade, double scale)
{
int i = 0;
double t = 0;
vector faces;
Mat gray, smallImg( cvRound (img.rows/scale), cvRound(img.cols/scale), CV_8UC1 );
cvtColor( img, gray, CV_BGR2GRAY );
resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
equalizeHist( smallImg, smallImg );
t = (double)cvGetTickCount();
cascade.detectMultiScale( smallImg, faces,
1.3, 2, CV_HAAR_SCALE_IMAGE,
Size(30, 30) );
t = (double)cvGetTickCount() - t;
printf( "detection time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
for( vector::const_iterator r = faces.begin(); r != faces.end(); r++, i++ )
{
IplImage* temp = cutImage(&(IplImage(img)), cvRect(r->x, r->y, r->width, r->height));
return temp;
}
return NULL;
}
//画直方图用
int HistogramBins = 256;
float HistogramRange1[2]={0,255};
float *HistogramRange[1]={&HistogramRange1[0]};
int CompareHist(IplImage* image1, IplImage* image2)
{
IplImage* srcImage;
IplImage* targetImage;
if (image1->nChannels != 1) {
srcImage = cvCreateImage(cvSize(image1->width, image1->height), image1->depth, 1);
cvCvtColor(image1, srcImage, CV_BGR2GRAY);
} else {
srcImage = image1;
}
if (image2->nChannels != 1) {
targetImage = cvCreateImage(cvSize(image2->width, image2->height), srcImage->depth, 1);
cvCvtColor(image2, targetImage, CV_BGR2GRAY);
} else {
targetImage = image2;
}
CvHistogram *Histogram1 = cvCreateHist(1, &HistogramBins, CV_HIST_ARRAY,HistogramRange);
CvHistogram *Histogram2 = cvCreateHist(1, &HistogramBins, CV_HIST_ARRAY,HistogramRange);
cvCalcHist(&srcImage, Histogram1);
cvCalcHist(&targetImage, Histogram2);
cvNormalizeHist(Histogram1, 1);
cvNormalizeHist(Histogram2, 1);
// CV_COMP_CHISQR,CV_COMP_BHATTACHARYYA这两种都可以用来做直方图的比较,值越小,说明图形越相似
printf("CV_COMP_CHISQR : %.4f\n", cvCompareHist(Histogram1, Histogram2, CV_COMP_CHISQR));
printf("CV_COMP_BHATTACHARYYA : %.4f\n", cvCompareHist(Histogram1, Histogram2, CV_COMP_BHATTACHARYYA));
// CV_COMP_CORREL, CV_COMP_INTERSECT这两种直方图的比较,值越大,说明图形越相似
printf("CV_COMP_CORREL : %.4f\n", cvCompareHist(Histogram1, Histogram2, CV_COMP_CORREL));
printf("CV_COMP_INTERSECT : %.4f\n", cvCompareHist(Histogram1, Histogram2, CV_COMP_INTERSECT));
cvReleaseHist(&Histogram1);
cvReleaseHist(&Histogram2);
if (image1->nChannels != 1) {
cvReleaseImage(&srcImage);
}
if (image2->nChannels != 1) {
cvReleaseImage(&targetImage);
}
return 0;
}
String srcImage = "d:\\ldh1.jpg";
String targetImage = "d:\\ldh5.jpg";
int main(int argc, char* argv[])
{
CascadeClassifier cascade;
namedWindow("image1");
namedWindow("image2");
if( !cascade.load( cascadeName ) )
{
return -1;
}
Mat srcImg, targetImg;
IplImage* faceImage1;
IplImage* faceImage2;
srcImg = imread(srcImage);
targetImg = imread(targetImage);
faceImage1 = detect(srcImg, cascade, 1);
if (faceImage1 == NULL) {
return -1;
}
// cvSaveImage("d:\\face.jpg", faceImage1, 0);
faceImage2 = detect(targetImg, cascade, 1);
if (faceImage2 == NULL) {
return -1;
}
// cvSaveImage("d:\\face1.jpg", faceImage2, 0);
imshow("image1", Mat(faceImage1));
imshow("image2", Mat(faceImage2));
CompareHist(faceImage1, faceImage2);
cvWaitKey(0);
cvReleaseImage(&faceImage1);
cvReleaseImage(&faceImage2);
return 0;
}