OpenCV3.2图像分割 实例5：GMM（高斯混合模型）图像分割

 1 #include <opencv2/opencv.hpp>
 2 #include <iostream>
 3  
 4 using namespace cv;
 5 using namespace cv::ml;
 6 using namespace std;
 7  
 8 int main(int argc, char** argv) {
 9     Mat src = imread("toux.jpg");
10     if (src.empty()) {
11         printf("could not load iamge...\n");
12         return -1;
13     }
14     namedWindow("input image", CV_WINDOW_AUTOSIZE);
15     imshow("input image", src);
16  
17     // 初始化
18     int numCluster = 3;
19     const Scalar colors[] = {
20         Scalar(255, 0, 0),
21         Scalar(0, 255, 0),
22         Scalar(0, 0, 255),
23         Scalar(255, 255, 0)
24     };
25  
26     int width = src.cols;
27     int height = src.rows;
28     int dims = src.channels();
29     int nsamples = width*height;
30     Mat points(nsamples, dims, CV_64FC1);
31     Mat labels;
32     Mat result = Mat::zeros(src.size(), CV_8UC3);
33  
34     // 图像RGB像素数据转换为样本数据 
35     int index = 0;
36     for (int row = 0; row < height; row++) {
37         for (int col = 0; col < width; col++) {
38             index = row*width + col;
39             Vec3b rgb = src.at<Vec3b>(row, col);
40             points.at<double>(index, 0) = static_cast<int>(rgb[0]);
41             points.at<double>(index, 1) = static_cast<int>(rgb[1]);
42             points.at<double>(index, 2) = static_cast<int>(rgb[2]);
43         }
44     }
45  
46     // EM Cluster Train
47     Ptr<EM> em_model = EM::create();
48     em_model->setClustersNumber(numCluster);
49     em_model->setCovarianceMatrixType(EM::COV_MAT_SPHERICAL);//设置协方差矩阵
50     //设置停止条件，训练100次结束
51     em_model->setTermCriteria(TermCriteria(TermCriteria::EPS + TermCriteria::COUNT, 100, 0.1));
52     em_model->trainEM(points, noArray(), labels, noArray());
53  
54     // 对每个像素标记颜色与显示
55     Mat sample(dims, 1, CV_64FC1);
56     double time = getTickCount();
57     int r = 0, g = 0, b = 0;
58     for (int row = 0; row < height; row++) {
59         for (int col = 0; col < width; col++) {
60             index = row*width + col;
61             int label = labels.at<int>(index, 0);
62             Scalar c = colors[label];
63             result.at<Vec3b>(row, col)[0] = c[0];
64             result.at<Vec3b>(row, col)[1] = c[1];
65             result.at<Vec3b>(row, col)[2] = c[2];
66  
67             /*b = src.at<Vec3b>(row, col)[0];
68             g = src.at<Vec3b>(row, col)[1];
69             r = src.at<Vec3b>(row, col)[2];
70             sample.at<double>(0) = b;
71             sample.at<double>(1) = g;
72             sample.at<double>(2) = r;
73             int response = cvRound(em_model->predict2(sample, noArray())[1]);
74             Scalar c = colors[response];
75             result.at<Vec3b>(row, col)[0] = c[0];
76             result.at<Vec3b>(row, col)[1] = c[1];
77             result.at<Vec3b>(row, col)[2] = c[2];*/
78  
79         }
80     }
81     printf("execution time(ms) : %.2f\n", (getTickCount() - time)/getTickFrequency()*1000);
82     imshow("EM-Segmentation", result);
83  
84     waitKey(0);
85     return 0;
86 }