OpenCV polar transform selective region

Question

I want to restrict the operating region of the polar transform in OpenCV's cvLogPolar function. I would consider rewriting the function from scratch. I am unwrapping a fisheye lens image to yield a panorama, and I want to make it as efficient as possible. Much of the image is cropped away after the transform, giving a donut-shaped region of interest in the input image:

fisheye lens image

This means much processing is wasted on black pixels.

This should be pretty simple, right? The function should take two additional arguments for clipping extents, radius1 and radius2. Here is the relevant pol-to-cart portion of the cvLogPolar function from imgwarp.cpp:

cvLogPolar( const CvArr* srcarr, CvArr* dstarr,
            CvPoint2D32f center, double M, int flags )
{
    cv::Ptr<CvMat> mapx, mapy;

    CvMat srcstub, *src = cvGetMat(srcarr, &srcstub);
    CvMat dststub, *dst = cvGetMat(dstarr, &dststub);
    CvSize ssize, dsize;

    if( !CV_ARE_TYPES_EQ( src, dst ))
        CV_Error( CV_StsUnmatchedFormats, "" );

    if( M <= 0 )
        CV_Error( CV_StsOutOfRange, "M should be >0" );

    ssize = cvGetMatSize(src);
    dsize = cvGetMatSize(dst);

    mapx = cvCreateMat( dsize.height, dsize.width, CV_32F );
    mapy = cvCreateMat( dsize.height, dsize.width, CV_32F );

    if( !(flags & CV_WARP_INVERSE_MAP) )
//---snip---
    else
    {
        int x, y;
        CvMat bufx, bufy, bufp, bufa;
        double ascale = ssize.height/(2*CV_PI);
        cv::AutoBuffer<float> _buf(4*dsize.width);
        float* buf = _buf;

        bufx = cvMat( 1, dsize.width, CV_32F, buf );
        bufy = cvMat( 1, dsize.width, CV_32F, buf + dsize.width );
        bufp = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*2 );
        bufa = cvMat( 1, dsize.width, CV_32F, buf + dsize.width*3 );

        for( x = 0; x < dsize.width; x++ )
            bufx.data.fl[x] = (float)x - center.x;

        for( y = 0; y < dsize.height; y++ )
        {
            float* mx = (float*)(mapx->data.ptr + y*mapx->step);
            float* my = (float*)(mapy->data.ptr + y*mapy->step);

            for( x = 0; x < dsize.width; x++ )
                bufy.data.fl[x] = (float)y - center.y;

#if 1
            cvCartToPolar( &bufx, &bufy, &bufp, &bufa );

            for( x = 0; x < dsize.width; x++ )
                bufp.data.fl[x] += 1.f;

            cvLog( &bufp, &bufp );

            for( x = 0; x < dsize.width; x++ )
            {
                double rho = bufp.data.fl[x]*M;
                double phi = bufa.data.fl[x]*ascale;

                mx[x] = (float)rho;
                my[x] = (float)phi;
            }
#else
//---snip---
#endif
        }
    }
    cvRemap( src, dst, mapx, mapy, flags, cvScalarAll(0) );
}

Since the routine works by iterating through pixels in the destination image, the r1 and r2 clipping region would just need to be translated to y1 and y2 row region. Then we just change the for loop: for( y = 0; y < dsize.height; y++ ) becomes for( y = y1; y < y2; y++ ). Correct?

What about constraining cvRemap? I am hoping it ignores unmoved pixels or it is a negligible computational cost.

Matt Montag · Accepted Answer · 2012-02-02 19:43:54Z

4

I ended up doing a different optimization: I store the result of the polar transform operation in persistent remapping matrices. This helps a LOT. If you're doing polar unwrap on full motion video using the same polar transform mapping at all times, you don't want to recalculate the transform with a million sin/cos operations every single frame. So this just required some small modification on the logPolar/linearPolar operations in the OpenCV source to save the remap result somewhere outside.

answered Feb 2, 2012 at 19:43

Matt Montag

7,6198 gold badges44 silver badges47 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Benoît Lahoz Over a year ago

It's an old post but I'm really (really) interested in you results ! Would you mind explaining your modifications on the lopPolar / linearPolar operations or... hum... post some code ? Thanks a lot !

Benoît Lahoz Over a year ago

Actually I'm trying to do the inverse : cartesian to polar coordinates, but I can't manage to use cartToPolar function...

Collectives™ on Stack Overflow

OpenCV polar transform selective region

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related