Search code examples
c++opencvimage-processingfiltering

How to perform OpenCV boxFilter for a ROI?

I have an image and I want to perform OpenCV boxFilter on a ROI of this image.

image = cv::imread(argv[1], CV_LOAD_IMAGE_COLOR);   
cv::Rect roi( 10, 10, 64, 64);
cv::Mat output;
cv::boxFilter(image(roi),output,-1,cv::Size(scale_size,scale_size));

I want to know what happen if the kernel is out of the ROI? Use the pixels out of the ROI but still within the image to do the filtering or just use the value specified by the BorderType?

Actually I want to use the former to do the filtering.

Solution

  • My original answer (got accepted) was that cv::boxFilter will not use the pixels out of the ROI.

    This actually turned out to be wrong.

    See @Micka's excellent answer that demonstrates that by default cv::boxFilter does use the pixels out of the ROI.
    It is also discussed in this github issue.

    I just wanted to add that although this behavior is not well documented, it is implied:

    cv::boxFilter has a flag-like parameter borderType.
    The values for it should come from enum cv::BorderTypes (some values can be combined by |). The last value of the enum is:

    BORDER_ISOLATED    = 16 //!< do not look outside of ROI

    This implies that if this value is not used, the function will look outside of the ROI.