根据矩阵的二维相关系数进行OCR识别

来源：互联网发布：沂南知豆租赁服务中心编辑：程序博客网时间：2024/06/04 19:12

我想通过简单的模板匹配来进行图像识别。

把预处理好的字符图片，分别与A到Z的样本图片进行模板匹配。结果最大的表明相关性最大，就可以识别字符图片了。

在实际应用中，我用了openCV的matchTemplate()函数，但是未达到我想要点的效果。

matchTemplate()的功能是在图像中搜索出指定的模板，如果模板是从待搜索的图像中截取出来的，会有很好的效果。但是如果模板不是待搜素图像的一部分，似乎达不到我想要的效果。

在尝试了matlab的corr2()后，发现corr2能很好的解决我的问题。

double Card::corr2(Mat matA, Mat matB){//计算两个相同大小矩阵的二维相关系数double corr2 = 0;double Amean2 = 0; double Bmean2 = 0;for (int m = 0; m < matA.rows; m++) {  uchar* dataA = matA.ptr<uchar>(m);  uchar* dataB = matB.ptr<uchar>(m); for (int n = 0; n < matA.cols;n++) {  Amean2 = Amean2 + dataA[n];  Bmean2 = Bmean2 + dataB[n];}                    }Amean2 = Amean2 / (matA.rows * matA.cols);Bmean2 = Bmean2 / (matB.rows * matB.cols);double Cov = 0;double Astd = 0;double Bstd = 0;for (int m = 0; m < matA.rows; m++) {  uchar* dataA = matA.ptr<uchar>(m);  uchar* dataB = matB.ptr<uchar>(m); for (int n = 0; n < matA.cols;n++) { //协方差Cov = Cov + (dataA[n] - Amean2) * (dataB[n] - Bmean2);//A的方差Astd = Astd + (dataA[n] - Amean2) * (dataA[n] - Amean2);//B的方差Bstd = Bstd + (dataB[n] - Bmean2) * (dataB[n] - Bmean2);}                    }corr2 = Cov / (sqrt(Astd * Bstd));return corr2;}

//待搜索图像Mat srcImage = imread("M:/图像处理实验/验证码/byx001.bmp",1);Mat resizeMat = Mat::zeros(25, 25, CV_8UC3);//缩放为25*25的矩阵。因为要相关匹配的模板大小为25*25resize(srcImage,resizeMat,resizeMat.size());//相关匹配double ccorrVal[26] = {0};double max = 0;int count = 0;for (int m = 0; m < 26; m++){char recogPath[100] = {1};strcpy(recogPath,"M://图像处理实验//验证码//大写字母//");char num[2] = {1};num[0] = 65 + m;strcat(recogPath, num);strcat(recogPath,".bmp");Mat img_display;resizeMat.copyTo( img_display );Mat std = imread(recogPath,0);Mat resizeMatSTD = Mat::zeros(25, 25, CV_8UC3);resize(std,resizeMatSTD,resizeMatSTD.size());adaptiveThreshold(resizeMatSTD ,resizeMatSTD ,255 ,ADAPTIVE_THRESH_MEAN_C ,THRESH_BINARY,5,1);corr2Val[m] = corr2(resizeMatSTD,img_display);if (max <= corr2Val[m]){max = corr2Val[m];count = m;}}char pathname[100]={1};strcpy(pathname,"M://图像处理实验//验证码//test//字符_");char num[10];_itoa(i, num, 10);strcat(pathname, num);char C[2] = {1};C[0] = 65 + count;strcat(pathname, C);strcat(pathname,".bmp");imwrite(pathname, resizeMat);

字符模板：

识别结果输出：

2 0