通过形态学操作实现图像边缘

实现图像边缘的算法的思路是：首先将图像转换为二值图像，然后膨胀此二值图像，最后减去未膨胀之前的图像便得到了图像的边缘。

参考代码如下所示：

int _tmain(int argc, _TCHAR* argv[])
{
    IplImage *srcImg = cvLoadImage("test.jpg");
    IplImage *grayImg = cvCreateImage(cvGetSize(srcImg), 8, 1);
    cvCvtColor(srcImg, grayImg, CV_BGR2GRAY);
    
    IplImage *dstImg = cvCreateImage(cvGetSize(srcImg), 8, 1);
	cvThreshold(grayImg, dstImg, cv::Otsu(srcImg), 255, CV_THRESH_BINARY);
    
    IplConvKernel *kernel = cvCreateStructuringElementEx(3, 3, 1, 1, CV_SHAPE_CROSS);

    cvMorphologyEx(dstImg, dstImg, NULL, kernel, CV_MOP_OPEN, 3);
    cvMorphologyEx(dstImg, dstImg, NULL, kernel, CV_MOP_CLOSE, 1);

    // Find edges: dilate(image) - image;
    IplImage *dilatedImg = cvCreateImage(cvGetSize(srcImg), 8, 1);
    cvDilate(dstImg, dilatedImg, kernel, 1);
    cvSub(dilatedImg, dstImg, dstImg);

    // Label edges in srcImg 
    IplImage *redCh = cvCreateImage(cvGetSize(srcImg), 8, 1);
    IplImage *greenCh = cvCreateImage(cvGetSize(srcImg), 8, 1);
    IplImage *blueCh = cvCreateImage(cvGetSize(srcImg), 8, 1);

    cvSplit(srcImg, blueCh, greenCh, redCh, NULL);
    cvOr(redCh, dstImg, redCh);
    cvMerge(blueCh, greenCh, redCh, NULL, srcImg);

    cvNamedWindow("src");
    cvShowImage("src", srcImg);
    cvNamedWindow("dst");
    cvShowImage("dst", dstImg);

    cvWaitKey(0);

    return 0;
}

其中cv::Otsu(srcImg)的算法，可参考http://en.wikipedia.org/wiki/Otsu%27s_method

输出结果：

关于Image Engineering & Computer Vision的更多讨论与交流，敬请关注本博客和新浪微博songzi_tea.