opencv不规则ROI——圆形ROI

A few days ago, a reader asked me about selecting circular ROI using OpenCV. I think such a function for selecting non-rectangular subimage would be useful, so I tried to write one.

 

OpenCV doesn't provide a function to make non-rectangle ROI, so I use image mask to do the trick. It works like this. say we have two images:

• img - an RGB image.
• mask - a black image, with white circle on it.

We can extract a region from img by:

res = img & mask

res now contains a portion from img where the white circle is located.

Fig 1. Selecting circular ROI and applying some processing on it.

The full listing is shown below. The code loads an image, select circular ROI, and applying some processing on it.

Listing 1: Extract a circular ROI

#include "cv.h"
#include "highgui.h"
 
int main(int argc, char** argv)
{
    IplImage* src, * res, * roi;
 
    /* usage: <prog_name> <image> */
    if (argc < 2)
    {
        fprintf(stderr, "Usage: %s <image>\n", argv[0]);
        return 1;
    }
 
    src = cvLoadImage(argv[1], 1);
    res = cvCreateImage(cvGetSize(src), 8, 3);
    roi = cvCreateImage(cvGetSize(src), 8, 1);
 
    /* prepare the 'ROI' image */
    cvZero(roi);
 
    /* Note that you can use any shape for the ROI */
    cvCircle(
        roi,
        cvPoint(130, 100),
        50,
        CV_RGB(255, 255, 255),
        -1, 8, 0
    );
 
    /* extract subimage */
    cvAnd(src, src, res, roi);
 
    /*
     * do the main processing with subimage here.
     * in this example, we simply invert the subimage
     */
    cvNot(res, res);
 
    /* 'restore' subimage */
    IplImage* roi_C3 = cvCreateImage(cvGetSize(src), 8, 3);
    cvMerge(roi, roi, roi, NULL, roi_C3);
    cvAnd(res, roi_C3, res, NULL);
 
    /* merge subimage with original image */
    cvNot(roi, roi);
    cvAdd(src, res, res, roi);
 
    /* show result */
    cvNamedWindow(argv[1], 1);
    cvNamedWindow("res", 1);
    cvShowImage(argv[1], src);
    cvShowImage("res", res);
 
    cvWaitKey(0);
 
    /* be tidy */
    cvDestroyAllWindows();
    cvReleaseImage(&src);
    cvReleaseImage(&res);
    cvReleaseImage(&roi);
 
    return 0;
}

　　In the example above, I only invert the subimage to make it simple. Change the line 39 with more complex image processing.