weird binary image values generated by cvThreshold

Hey, guys, i am using opencv to do some vehicle recognition work, and when i use cvThershold to convert the gray image to binary image, th开发者_如何学编程e return image is really strange, the binary image supposes to have only two values,0 and 255, however, it contains other values like 2,3,254,253, anyone knows how this happens, and also cvCmps also has this problem.

---

cvThreshold has a variety of behaviours beyond normal binary thresholding. They are described in the OpenCV API reference.

For example, if you call it with the flag threshold_type set CV_THRESH_TRUNC, it will truncate all intensities above the specified threshold only. The intensities below the threshold will remain untouched. Perhaps this accounts for your strange result?

If you post the image and your code (the bit that calls cvThreshold is enough) I could probably be of more help.

---

Try this:

/*
 * compile with:
 *
 * g++ -Wall -ggdb -I. -I/usr/include/opencv -L /usr/lib -lm -lcv -lhighgui -lcvaux threshold.cpp -o threshold.out
 */

#include <cv.h>    
#include <highgui.h>
#include <stdio.h>
#include <assert.h>

IplImage *
threshold(IplImage const *in, int threshold)
{
    assert(in->nChannels == 1);
    CvSize size = cvSize(in->width, in->height);
    IplImage *out = cvCreateImage(size, IPL_DEPTH_8U, 1);
    cvThreshold(in, out, threshold, 255, CV_THRESH_BINARY);

    return out;
}

void
show_image(char const *title, IplImage const *image)
{
    cvNamedWindow(title, CV_WINDOW_AUTOSIZE);
    cvShowImage(title, image);
    cvWaitKey(0);    
    cvDestroyWindow(title);
}

int
main(int argc, char **argv)
{
    if (argc < 2)
    {
        fprintf(stderr, "usage: %s in.png\n", argv[0]);
        return 1;
    }

    IplImage *in = cvLoadImage(argv[1]);
    IplImage *grey = in;

    if (in->nChannels != 1)
    {
        /*
         * For some reason, cvCreateImage returns an image with 3 channels even
         * when a greyscale image is specified (e.g. PGM).  Hack around this by
         * just taking the first channel of the image.  OpenCV uses BGR order,
         * so it will be the B channel.
         */
        CvSize size = cvSize(in->width, in->height);
        grey = cvCreateImage(size, IPL_DEPTH_8U, 1);
        cvSplit(in, grey, NULL, NULL, NULL);
        cvReleaseImage(&in);
    }

    IplImage *thres = threshold(grey, 127);
    show_image("thresholded", thres);
    cvReleaseImage(&thres);

    cvReleaseImage(&grey);
    return 0;
}

Give it any image (even a colour one, see comment for clarification), e.g. [this one][1]. Do you get the expected result?

[1]: http://r0k.us/graphics/kodak/kodak/kodim20.png SixShooter