OpenCV: how to set alpha transparency of a pixel

Question

I have an image I am trying to segment by colouring each pixel either red green or blue. I have calculated a confidence score for each pixel and want to adjust the alpha transparency of the pixel to reflect confidence, i.e. low confidence means almost transparent. Is there a way to do this in OpenCV? If not can anyone suggest a minimally invasive library (C++)?

I have tries using a 4 channel 8-bit Mat as suggested by Aurellius, here is the code:

cv::Mat m = cv::Mat(20,20, CV_8UC4);
    for(int i = 0; i < m.rows; i++){
        for(int j = 0; j < m.cols; j++){
            Vec4b& v = m.at<Vec4b>(i,j);
            v[0] = 255;
            v[1] = 0;
            v[2] = 0;
            v[3] = 0.5;
        }
    }

    imwrite("alpha.png", m);
    namedWindow("m");
    imshow("m", m);
    waitKey(0);

The image shown is just all blue (no transparency) and the image just fully transparent.

Answer 1

There are many ways to accomplish this. One possible way is to access and modify each individual pixel. Assuming image is a four-channel, 8-bit cv::Mat:

auto& pixel = image.at<cv::Vec4b>(i,j);
pixel[3] = confidence;

where i and j are the indices of the pixel in the image.

There are other methods that are probably more elegant, but they will depend on your current code.

UPDATE: The behavior you describe is to be expected. Apparently cv::imshow() does not support transparency. This explains why your displayed image is all blue.

As for the saved image, it is important to remember that the image is of type CV_8UC4. That means that each channel element is stored as a uchar. Assigning a value of 0.5 will truncate to zero, hence the fully transparent saved image.

If your confidence is a float value in the range [0,1], scale it by 255 to put it in the range supported by 8-bit images. Thus,

v[3] = 0.5;

becomes

v[3] = 0.5 * 255;