Qimage to cv::Mat convertion strange behaviour

Question

I am trying to create an application where I am trying to integrate opencv and qt.

I managed successfully to convert a cv::Mat to QImage by using the code below:

void MainWindow::loadFile(const QString &fileName)
{
    cv::Mat tmpImage = cv::imread(fileName.toAscii().data());
    cv::Mat image;

    if(!tmpImage.data || tmpImage.empty())
    {
        QMessageBox::warning(this, tr("Error Occured"), tr("Problem loading file"), QMessageBox::Ok);
        return;
    }

/* Mat to Qimage */
    cv::cvtColor(tmpImage, image, CV_BGR2RGB);
    img = QImage((const unsigned char*)(image.data), image.cols, image.rows, QImage::Format_RGB888);

    imgLabel->setPixmap(QPixmap::fromImage(img));
    imgLabel->resize(imgLabel->pixmap()->size());

    saveAsAct->setEnabled(true);
}

However, when I am trying to convert the QImage to cv::Mat by using the following code:

bool MainWindow::saveAs()
{
    if(fileName.isEmpty())
    {
        QMessageBox::warning(this, tr("Error Occured"), tr("Problem loading file"), QMessageBox::Close);
        return EXIT_FAILURE;
    }else{
        outputFileName = QFileDialog::getSaveFileName(this, tr("Save As"), fileName.toAscii().data(), tr("Image Files (*.png *.jpg *.jpeg *.bmp)\n *.png\n *.jpg\n *.jpeg\n *.bmp"));

    /* Qimage to Mat */
    cv::Mat mat = cv::Mat(img.height(), img.width(), CV_8UC4, (uchar*)img.bits(), img.bytesPerLine());
    cv::Mat mat2 = cv::Mat(mat.rows, mat.cols, CV_8UC3 );
    int from_to[] = {0,0,  1,1,  2,2};
    cv::mixChannels(&mat, 1, &mat2, 1, from_to, 3);

    cv::imwrite(outputFileName.toAscii().data(), mat);
}

saveAct->setEnabled(true);
return EXIT_SUCCESS;
}

I have no success and the result is totally disordered image. In the net that I searched I saw that the people are using this way without mentioning any specific problems. Does someone have any idea, about what could be cause the problem? Thanks in advance.

Theoodore

P.S. I am using opencv 2.4 and Qt 4.8, under a Arch Linux system with gnome-3.4

Adequacy

Answer 1

Just find out the "correct" solution of copying(not reference) the QImage to cv::Mat

The answer of Martin Beckett is almost correct

for (int i=0;i<image.height();i++) {
    memcpy(mat.ptr(i),image.scanline(i),image.bytesperline());
}

I don't see the full codes, but I guess you may want to use it like this way

cv::Mat mat(image.height(), image.width(), CV_8UC3);
for (int i=0;i<image.height();i++) {
        memcpy(mat.ptr(i),image.scanline(i),image.bytesperline());
    }

But this code exist a problem, the memory allocated by cv::Mat may not have the same "bytesperline" as the QImage

The solution I found is take the reference of the QImage first, then clone it

return cv::Mat(img.height(), img.width(), format, img.bits(), img.bytesPerLine()).clone();    

The solution of Martin Beckett suggested could generate correct result in most of the times, I didn't notice there are a bug until I hit it.

Whatever, I hope this is a "correct" solution. If you find any bug(s), please let everyone know so we could have a change to improve the codes.

Decurved

Answer 2

opencv images are stepped so that each row begins on a multiple of 32bits, this makes memory access faster. If you are using a 3byte/pixel format then unless you have a width that is 1/3 of a multiple of 4 you will have 'spare' memory at the end of each line

The safe way is to copy the image data a row at a time. The opencv mat.ptr(row) returns a pointer to the start of each row and the QImage .scanline(row) member does the same

See How to output this 24 bit image in Qt

edit: Something like

for (int i=0;i<image.height();i++) {
    memcpy(mat.ptr(i),image.scanline(i),image.bytesperline());
}

Scone

i changed the code for "cv::Mat to QImage" to cv::cvtColor(tmpImage, image, CV_BGR2RGB); uchar *buffer = image.ptr(); img = QImage(buffer, image.cols, image.rows, image.step, QImage::Format_RGB888); imgLabel->setPixmap(QPixmap::fromImage(img)); imgLabel->resize(imgLabel->pixmap()->size()); but it does not seem to have any affect. The problem i think is how to convert the "QImage to cv::Mat" correctly. – Adequacy

No you have to loop over all the rows copying one at a time, because they aren't necessarily contiguous in one of the formats, see the code in the link – Scone

Answer 3

I think you might find this useful. http://www.jdxyw.com/?p=1480

It uses IplImage to get data, but you can use cv::Mat and Mat.data to get a pointer to the original matrix. I hope you might find this useful.

Caucasian

Answer 4

From this source code

QImage MatToQImage(const Mat& mat)
{
    // 8-bits unsigned, NO. OF CHANNELS=1
    if(mat.type()==CV_8UC1)
    {
        // Set the color table (used to translate colour indexes to qRgb values)
        QVector<QRgb> colorTable;
        for (int i=0; i<256; i++)
            colorTable.push_back(qRgb(i,i,i));
        // Copy input Mat
        const uchar *qImageBuffer = (const uchar*)mat.data;
        // Create QImage with same dimensions as input Mat
        QImage img(qImageBuffer, mat.cols, mat.rows, mat.step, QImage::Format_Indexed8);
        img.setColorTable(colorTable);
        return img;
    }
    // 8-bits unsigned, NO. OF CHANNELS=3
    if(mat.type()==CV_8UC3)
    {
        // Copy input Mat
        const uchar *qImageBuffer = (const uchar*)mat.data;
        // Create QImage with same dimensions as input Mat
        QImage img(qImageBuffer, mat.cols, mat.rows, mat.step, QImage::Format_RGB888);
        return img.rgbSwapped();
    }
    else
    {
        qDebug() << "ERROR: Mat could not be converted to QImage.";
        return QImage();
    }
} // MatToQImage()

Octant

OP wants QImage to Mat – Span

Answer 5

This is what i got from here thanks to jose. It helps me to get over this.

To visualize OpenCV image (cv::Mat) in Qt (QImage), you must follow these steps:

1. Invert color sequence: cv::cvtColor(imageBGR, imageRGB, CV_BGR2RGB);
2. Change format OpenCV to Qt: QImage qImage((uchar*) imageRGB.data, imageRGB.cols, imageRGB.rows, imageRGB.step, QImage::Format_RGB888);
3. Use QPainter to render the image.

Please note the use of QImage::Format. Read the Qt on this issue.

Radbourne

OP wants QImage to cv::Mat – Span

Answer 6

if you're still looking for the solution. here's one :

Cv::Mat to QImage :

QImage Mat2QImage(cv::Mat &image )
{
    QImage qtImg;
    if( !image.empty() && image.depth() == CV_8U ){
        if(image.channels() == 1){
            qtImg = QImage( (const unsigned char *)(image.data),
                            image.cols,
                            image.rows,
                            QImage::Format_Indexed8 );
        }
        else{
            cvtColor( image, image, CV_BGR2RGB );
            qtImg = QImage( (const unsigned char *)(image.data),
                            image.cols,
                            image.rows,
                            QImage::Format_RGB888 );
        }
    }
    return qtImg; 
}

For QImage to cv::Mat .

cv::Mat QImage2Mat(QImage &image) {

cv::Mat cvImage;
  switch (image.format()){
  case QImage::Format_RGB888:{
      cvImage = cv::Mat(image.height(),
                        image.width(),
                        CV_8UC3,
                        image.bits(),
                        image.bytesPerLine());
      cv::cvtColor(cvImage, cvImage, CV_RGB2BGR);
      return cvImage;
  }
  case QImage::Format_Indexed8:{
      cvImage = cv::Mat(image.height(),
                        image.width(),
                        CV_8U,
                        image.bits(),
                        image.bytesPerLine());
      return cvImage;
  }
  default:
      break;
  }
  return cvImage;
}

Warrigal