将OpenCV Mat 32FC1数据复制到WriteableBitmap

Question

The very short story is that I want to copy a CV_32FC1 OpenCV image to a Windows [WriteableBitmap][1] which is formatted as [PixelFormats.Gray32Float][2]

The following code works for all the 8 bit image formats (CV_8UF1, CV_8UF3 etc)

writeableBitmap = new WriteableBitmap(imgWrapper.Width, 
                                      imgWrapper.Height, 96, 96,
                                      imgWrapper.PixelFormat, null);

var frameSize = new Int32Rect(0, 0, imgWrapper.Width, imgWrapper.Height);
writeableBitmap.WritePixels(frameSize, imgWrapper.Data, imgWrapper.Stride, 0);

imgWrapper.Data is an array byte[], so I am "blindly" copying the data to the WriteableBitmap. Apparently the WriteableBitmap and the OpenCV images have the same layout for 8 bit images, but maybe not for gray 32 bit images.

1) How does OpenCV format 32FC1 Mat data? 2) Is there a way in Visual Studio to see the Mat.data as 32 bit floats. In the debugger it's always shown as bytes. 3) I've read about scaling 32F images to 0-1. Is this universal? Does the Windows WriteableBitmap have the same interpretation of the data?

Thanks for any help or tips

Answer 1

To see the value for each pixel in an CV_32FC1 Mat, I ended up dumping it to an xml file.

string filename = "averageImage.xml";
FileStorage fs(filename, FileStorage::WRITE);
fs << "R" << averageImage32FNormalized; 

To see the data within VS, you can point the Mat data to a float array float* data = (float*)(averageImage32F.data); Then type "data,100" in the watch window (for example to see the first 100 values)

For display in WPF, the floating point values need to be scaled to 0-1

Mat averageImage32FNormalized;
cv::normalize(averageImage32F, averageImage32FNormalized, 0, 1, NORM_MINMAX);

Then you can use writeableBitmap.WritePixels as shown in my original post.

Note that cv::normalize as I've shown above doesn't normalize the darkest pixel to 0 and the lightest pixel to 1. It seems to assume that the floating point values are on 0-255 scale. In other words a floating point value of 150.3 was converted to 0.59 (150.3/255).