OpenCV：无法使用k-means进行图像分割

Question

Before going into deep of my question, I want you to know that I've read other posts on this forum, but none regards my problem. In particular, the post here answers the question "how to do this?" with k-means, while I already know that I have to use it and I'd like to know why my implementation doesn't work.

I want to use k-means algorithm to divide pixels of an input image into clusters, according to their color. Then, after completing such task, I want each pixel to have the color of the center of the cluster it's been assigned to. Taking as reference the OpenCV examples and other stuff retrieved on the web, I've designed the following code:

#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
#include <iostream>

using namespace std;
using namespace cv;


int main( int argc, char** argv )
{

    Mat src = imread( argv[1], 1 );

    // reshape matrix
    Mat resized(src.rows*src.cols, 3, CV_8U);
    int row_counter = 0;

    for(int i = 0; i<src.rows; i++)
    {
        for(int j = 0; j<src.cols; j++)
        {
            Vec3b channels = src.at<Vec3b>(i,j);
            resized.at<char>(row_counter,0) = channels(0);
            resized.at<char>(row_counter,1) = channels(1);
            resized.at<char>(row_counter,2) = channels(2);

            row_counter++;
        }
    }

    //cout << src << endl;

    // change data type
    resized.convertTo(resized, CV_32F);

    // determine termination criteria and number of clusters
    TermCriteria criteria(TermCriteria::COUNT + TermCriteria::EPS, 10, 1.0);
    int K = 8;

    // apply k-means
    Mat labels, centers;
    double compactness = kmeans(resized, K, labels, criteria, 10, KMEANS_RANDOM_CENTERS, centers);

    // change data type in centers
    centers.convertTo(centers, CV_8U);

    // create output matrix
    Mat result = Mat::zeros(src.rows, src.cols, CV_8UC3);

    row_counter = 0;
    int matrix_row_counter = 0;
    while(row_counter < result.rows)
    {
        for(int z = 0; z<result.cols; z++)
        {
            int index = labels.at<char>(row_counter+z, 0);
            //cout << index << endl;
            Vec3b center_channels(centers.at<char>(index,0),centers.at<char>(index,1), centers.at<char>(index,2));
            result.at<Vec3b>(matrix_row_counter, z) = center_channels;
        }

        row_counter += result.cols;
        matrix_row_counter++;
    }

    cout << "Labels " << labels.rows << " " << labels.cols << endl;
    //cvtColor( src, gray, CV_BGR2GRAY );
    //gray.convertTo(gray, CV_32F);

    imshow("Result", result);
    waitKey(0);


    return 0;
}

Anyway, at the end of computation, I simply get a black image. Do you know why? Strangely, if I initialize result matrix as

Mat result(src.size(), src.type())

at the end of algorithm it will display exactly the input image, without any segmentation.

In particular, I have two doubts:

1) is it correct to lay the RGB values of a pixel on each row of matrix resized the way I've done it? is there a way to do it without a loop?

2) what's exactly the content of centers , after k-means function finishes working? it's a 3 columns matrix, does it contains the RGB values of clusters' centers?

thanks for support.

Answer 1

Mat& ScanImageAndReduceC(Mat& I) {

// accept only char type matrices CV_Assert(I.depth() == CV_8U);

// accept only char type matrices CV_Assert(I.depth() == CV_8U);

 int channels = I.channels(); int nRows = I.rows; int nCols = I.cols * channels; if (I.isContinuous()) { nCols *= nRows; nRows = 1; } int i, j; uchar* p; for (i = 0; i < nRows; ++i) { p = I.ptr<uchar>(i); for (j = 0; j < nCols; ++j) { I.at<uchar>(10, 10) = 255; } } return I; 

}

-------Main Program-------

 diff = ScanImageAndReduceC(diff); \n\nnamedWindow("Difference", WINDOW_AUTOSIZE);// Create a window for display. \nimshow("Difference", diff);  // Show our image inside it. \n\nwaitKey(0);  // Wait for a keystroke in the window \nreturn 0; \n

}