why my imshow differs from imwrite

Question

I am using OpenCV 2.4.8 on VS2013. I have the following snippet of code:

    Mat img_sum = Mat::zeros(img_gray.size(), CV_32F);
    char file_name[FILENAME_MAX];
    for (int i = 0; i < SCALE_SIZE; i++){
        for (int j = 0; j < ORIENTATION_SIZE; j++){
            Mat srcImg;
            g_gabor[i*ORIENTATION_SIZE + j].conv_img(img_gray, srcImg, CV_GABOR_REAL);

            memset(file_name, NULL, FILENAME_MAX);
            sprintf_s(file_name, "gabor_images/%d_%d.png", i, j);
            imwrite(file_name, srcImg);
            imshow("Gabor滤波结果", srcImg);    //此时srcImg.type() == 1
            waitKey(100);
        }
    }
    imwrite("img_sum.png", img_sum);

The key issue is result of imwrite and imshow. They gave me different result.

The imshow image: imshow image

And, the imwrite image: imwrite image

I was wonderring if the reason is the image type. If so, how to convert image type to solve the problem.

imshow will render an image that can be displayed by your monitor. typically that is a 24 bit RGB image. On image writing to disk, typically 8 bit channel information is used, too. So if you have a 32 bit float matrix, opencv has to decide on how to render or save it. From documentation you'll see that for rendering, <= 0 will be rendered as black and >= 1 will be rendered as white, while for disk saving with imwrite, no scaling will apply so you'll have to scale manually before calling imwrite. — Micka, Aug 03 '16 at 08:45

score 7 · Answer 1 · answered Aug 03 '16 at 08:35

7

With type CV_32F, you have to multiply your image by 255 before using imwrite.

Try imwrite(file_name, 255 *srcImg);.

answered Aug 03 '16 at 08:35

Sunreef

4,452
21
33

Thanks for your answer! My Image Type here is equal to 1, not CV_32F. Does it still works? – Frank Aug 03 '16 at 08:46
@Frank Can you show how you define your `img_gray` ? It seems strange that your `srcImg` should be of type `CV_8S` – Sunreef Aug 03 '16 at 09:04
You are right. Now I solve my problem by function convertTo(dst, type, scale, shift). – Frank Aug 03 '16 at 09:09

score 1 · Accepted Answer · answered Aug 03 '16 at 08:37

1

When you convert to CV_8U first, for instance with convertTo(dst, type, scale, shift), you can be sure both are identical.

answered Aug 03 '16 at 08:37

SpamBot

1,438
11
28

score 0 · Answer 3 · answered Aug 03 '16 at 09:06

Now I solve my question by OpenCV function convertTo. It is not only a problem of the TYPE, scale and shift are needed, too.

        double minVal, maxVal;
        minMaxLoc(srcImg, &minVal, &maxVal); //find minimum and maximum intensities
        Mat charImage;
        srcImg.convertTo(charImage, CV_8U, 255.0 / (maxVal - minVal), -minVal);
        imwrite(file_name, charImage);
        imshow("Gabor滤波结果", srcImg);    //此时srcImg.type() == 1

why my imshow differs from imwrite

3 Answers3

Linked