Get RGB value of pixel

J.Hilk

for (int row=0; row<img.height(); row++) {
for (int col=0; col<img.width(); col++) {
     QRgb pixColor = img.pixel(row, col);
     qDebug() << "red" << qRed(pixColor) << endl
                       << "green" <<qGreen(pixColor) << endl
                       << "blue" << qBlue(pixColor);
  }
}

however, QImage::pixel is slow the docu recommends constBits() or constScanLine()

MHermann

@J.Hilk : Thanks for your answers.
But this also gives me the same results.
I thought maybe the weird results are because of the image format and/or the color table.

If I am using Indexed8 format, I also set the greated color table. Then I receive one value for blue. But red and green are always 0.
If I am using Rgb32 format, I do not set the color table. Then I receive the same values for red, green and blue.

mranger90

I think the problem is your color table. The color table should be something like:

  for (int i = 0; i < 256; i++) {
        QRgb color(qRgba(i,i,i,0xFF));
        m_colors.append(color);
    }

My example sets up a 255 entry gray scale table. You may want something different.
The point is, there should only be 255 entries.

MHermann

@mranger90 : Thanks for your answer.
I want to detect red pixels in the QImage.
How does the color have to look like if I want to detect also colors, not only different gray values?

If I am using a Indexed8 image I can only detect gray values so I have to convert the image to RGB32, right?

mranger90

By using an indexed-8 format you are creating an image based on 8 bit data. That means there are only 256 possible values for the pixel. By setting up a color table your are telling QImage how to map those 8 bit values to the RGB values. It may or may not be gray scale. You could set up a table in which the values map to color based RGB values.
The point is, you are setting up an enormous table (256256256) and that cant be correct.
Once the table is correct, you should be able extract the color for a given pixel as mentioned by @J-Hilk

MHermann

@mranger90 : Thanks for your hints.
Now i set up a color table with 256 entries with different colors.
Now it seems to work.

MHermann

@mranger90 : After some testing I think it is not working correctly...
I get some values. But theses values are not representing the real colors of my image.

I built up the color table with 256 color values, like this:

   for (int colCnt=2, cnt=0; cnt<41; cnt++) {
        m_colorTable.append(qRgba(   174, colCnt,      0, 0xFF)); // Rot  -> Gelb
        m_colorTable.append(qRgba(colCnt,    174,      0, 0xFF)); // Grün -> Gelb
        m_colorTable.append(qRgba(     0,    174, colCnt, 0xFF)); // Grün -> Türkis
        m_colorTable.append(qRgba(colCnt,      0,    174, 0xFF)); // Blau -> Magenta
        m_colorTable.append(qRgba(     0, colCnt,    174, 0xFF)); // Blau -> Türkis
        m_colorTable.append(qRgba(   174,      0, colCnt, 0xFF)); // Rot  -> Magenta
        colCnt+=2;
    }

    m_colorTable.append(qRgba(255,   0,   0, 0xFF)); // Rot
    m_colorTable.append(qRgba(255, 255,   0, 0xFF)); // Gelb
    m_colorTable.append(qRgba(255,   0, 255, 0xFF)); // Magenta
    m_colorTable.append(qRgba(  0, 255,   0, 0xFF)); // Grün
    m_colorTable.append(qRgba(  0, 255, 255, 0xFF)); // Türkis
    m_colorTable.append(qRgba(  0,   0, 255, 0xFF)); // Blau
    m_colorTable.append(qRgba(100, 100, 100, 0xFF)); // Dunkelgrau
    m_colorTable.append(qRgba(150, 150, 150, 0xFF)); // Hellgrau
    m_colorTable.append(qRgba(255, 255, 255, 0xFF)); // Weiss
    m_colorTable.append(qRgba(  0,   0,   0, 0xFF)); // Schwarz

What colors do I have to use in the color table? Is there a rule that I can follow?
At the moment I searched for color tables with 256 values in the internet and found the values I used in the code above. But these values seem not to work ...
Can anyone explain me how I can know which colors I have to add to the color table?

Btw.: My goal is to detect red color in this image.

ollarch

Hi,

Why are you using a 8 bit Indexed image? You can use the RGB Image format instead. The 8 bit Indexed image can only represent 256 color values and may not be the values you want.

"If I am using Rgb32 format, I do not set the color table. Then I receive the same values for red, green and blue."

This is the way to go.

MHermann

@ollarch : Thanks for your answer.
In meanwhile I found out that the QVideoFrame comes in format YUV420P.
I think I have to convert this frame to a RGB image.
But at the moment this is not yet working properly...

ollarch

@MHermann said in Get RGB value of pixel:

YUV420P

Well, if you want to get the pixel value on some position in the image you could get the pointer to the image bits "uchar* imageBits()" to get the bits of the internal data and then do the YUV to RGB conversion. This will be faster instead of converting the entire image to RGB, so it depends on what you want to achieve.

MHermann

@ollarch : Thanks.
But at the moment I have the problem that I don't have a valid image.
I get a QVideoFrame in format QVideoFrame::Format_YUV420P.
I think I first have to convert this QVideoFrame to a QImage. And after that I want to search for red pixels in this image.
What is your advice? Can you tell me how I can get a RGB image? Or is it possible to get a YUV image and then search there for the red pixels?

ollarch

@MHermann The easiest way to search for red pixels would be to convert the image to RGB first. I don't know how to do the conversion but you can search for this conversion on your search engine. It will not be so difficult to get the RGB pixels from a YUV pixels. You also can look how OpenCV does the conversion.

Finally, for searching the red pixels, do you want only the pixels with R=255,G=0,B=0 ?

MHermann

@ollarch : Ok. At the moment I try to do the conversion correctly.
I am not sure if I want to search only for 255,0,0. Does this matter? Or do you mean that I can use V component of YUV then?

ollarch

@MHermann I was only asking. I don't know how to work directly with the YUV components. I was thinking that getting only the (255,0,0) values on a real image will get you 0 pixels as you will not get a perfect red pixels on an image(could be but it will be strange).

MHermann

@ollarch : Ok. Now it works. I am doing the conversion from YUV to RGB. And then I am able to get the RGB values of each pixel.

ollarch

@MHermann now that you have a solution would be great to share your solution with other people to help them if they need to implement the same.

MHermann

@ollarch : Of course. Maybe someone is interested in this:

QImage convertFrameYUV420ToImageRGB888(QVideoFrame vidFrame)
{
    vidFrame.map(QAbstractVideoBuffer::ReadOnly);

    unsigned char* pData = (unsigned char *) vidFrame.bits();
    unsigned char pDataRgb[vidFrame.width()*vidFrame.height()*3];

    int bytesPerLine = vidFrame.bytesPerLine();
    int numYBytes = vidFrame.width()*vidFrame.height();
    int numUBytes = numYBytes / 4;
    int idxY = 0;
    int idxU = numYBytes;
    int idxV = numYBytes+numUBytes;
    int dstIdx=0;
    for (int h=0; h<vidFrame.height(); h++) {
        for (int w=0; w<vidFrame.width(); w++) {

            idxY = h*bytesPerLine+w;
            int y = pData[idxY];

            idxU = (int)(numYBytes+(h/2)*(bytesPerLine/2)+w/2);
            int u = pData[idxU];

            idxV = (int)(numYBytes*1.25 + (h/2)*(bytesPerLine/2)+w/2);
            int v = pData[idxV];

            pDataRgb[dstIdx++] = y + 1.402 * (v-128);                   // r
            pDataRgb[dstIdx++] = y - 0.344 * (u-128) - 0.714 * (v-128); // g
            pDataRgb[dstIdx++] = y + 1.772 * (u-128);                   // b
        }
    }

    vidFrame.unmap();

    QImage imgRgb (pDataRgb, vidFrame.width(), vidFrame.height(), QImage::Format_RGB888);
    return imgRgb;
}

And after that conversion go over all pixels in the image and access them via QImage::PixelColor().

ollarch

@MHermann Thank you for sharing. Now, I can suggest you some modifications to make it faster.
First of all, the "vidFrame" could be changed to a pointer or a const reference to avoid the copy of the QVideoFrame object(time consuming).
Finally, reading the Qt Docs, the "pixelColor" method "is expensive when used for massive pixel manipulations. Use constBits() or constScanLine() when many pixels needs to be read". I suggest you to access directly to the internal data bits and check the values there using data pointer offsets.

MHermann

@ollarch : Thanks for your suggestions.
But I think I have to copy the QVideoFrame. Because I can only access the bytes of the frame while they are mapped. I am doing this conversion and further code in a separate thread. So if I map and then unmap the frame the bytes are no longer accessable.