How to use cvTransform?

[Vladimir Eremeev]

It seems I am missing some basics, however, I haven't found any info on the net.

I have an affine transformation matrix which I am going to use to transform images with cvWarpAffine.
I have an image to warp, and I need to esimate the resulting image size in order to correctly allocate the memory.

My method is the following:
Estimate new coordinates for the corners and determine the surrounding rectangle.

I have created the matrix with the image corners' coordinates
/0 , 0\
|w, 0|
|0,  h|
\w, h/
where w and h are the width and height of the source image.

Then I was going to feed it to cvTransform, find a maximum and minimum values for each coordinate and estimate new width and height.

However, cvTransform was returning errors.

The failed C++ code is below

    IplImage *src=cvLoadImage("images/2/1.JPG");

// create initial matrix.
    CvMat *corners=cvCreateMat(4,2,CV_64FC1);
// fill it
    cvmSet(corners,0,0,0.);                      cvmSet(corners,0,1,0.);                              //(0,0)
    cvmSet(corners,1,0,double(src->width)); cvmSet(corners,1,1,0.);                              //(w,0)
    cvmSet(corners,2,0,0.);      cvmSet(corners,2,1,double(src->height));                      //(0,h)
    cvmSet(corners,3,0,double(src->width)); svmSet(corners,3,1,double(src->height));     //(w,h)

    CvMat *corners_tr=cvCreateMat(4,2,CV_64FC1);

// and calculate the transformation
    cvTransform(corners,corners_tr,res);

// then find min and max
    int minh=cvmGet(corners_tr,0,0);
    int maxh=minh,minw=minh,maxw=minh;

    for(i=0;i<4;i++){
    int c=cvmGet(corners_tr,0,i);
    if(maxw < c) maxw=c;
    if(minw > c) minw=c;
    c=cvmGet(corners_tr,1,i);
    if(maxh < c) maxh=c;
    if(minh > c) minh=c;
    }

// and calculate dst_size
    CvSize dst_size=cvSize(maxw-minw,maxh-minh);

I tried transposing the matrix 'corners' or adding a new row with 1s - no success.

[shanaka]

Vladimir Eremeev wrote:

It seems I am missing some basics, however, I haven't found any info on the net.

I have an affine transformation matrix which I am going to use to transform images with cvWarpAffine.
I have an image to warp, and I need to esimate the resulting image size in order to correctly allocate the memory.

My method is the following:
Estimate new coordinates for the corners and determine the surrounding rectangle.

I have created the matrix with the image corners' coordinates
/0 , 0\
|w, 0|
|0,  h|
\w, h/
where w and h are the width and height of the source image.

Then I was going to feed it to cvTransform, find a maximum and minimum values for each coordinate and estimate new width and height.

However, cvTransform was returning errors.

The failed C++ code is below

    IplImage *src=cvLoadImage("images/2/1.JPG");

// create initial matrix.
    CvMat *corners=cvCreateMat(4,2,CV_64FC1);
// fill it
    cvmSet(corners,0,0,0.);                      cvmSet(corners,0,1,0.);                              //(0,0)
    cvmSet(corners,1,0,double(src->width)); cvmSet(corners,1,1,0.);                              //(w,0)
    cvmSet(corners,2,0,0.);      cvmSet(corners,2,1,double(src->height));                      //(0,h)
    cvmSet(corners,3,0,double(src->width)); svmSet(corners,3,1,double(src->height));     //(w,h)

    CvMat *corners_tr=cvCreateMat(4,2,CV_64FC1);

// and calculate the transformation
    cvTransform(corners,corners_tr,res);

// then find min and max
    int minh=cvmGet(corners_tr,0,0);
    int maxh=minh,minw=minh,maxw=minh;

    for(i=0;i<4;i++){
    int c=cvmGet(corners_tr,0,i);
    if(maxw < c) maxw=c;
    if(minw > c) minw=c;
    c=cvmGet(corners_tr,1,i);
    if(maxh < c) maxh=c;
    if(minh > c) minh=c;
    }

// and calculate dst_size
    CvSize dst_size=cvSize(maxw-minw,maxh-minh);

I tried transposing the matrix 'corners' or adding a new row with 1s - no success.

I think ur roblom is that u r using 4x2 vector where as u should be using 4x1 3-channel vector with (x,y,1) cordinates :)

I ve used this function (for perspective transformation using a homography to map form one image to another) but for me it gives unexpected values. and when i used cvPerspectiveTransform() it all the transformed cordinates as 0,0.
i've tried to manually transform one point using cvGEMM() function acording to the equation p' = H x p and it works fine but when i give the data to this function it gives wrong cordinates..... is there any idea about it? 

[Odahlam]

I've been confronted with this exact same problem and finally found a solution:
Use cvProjectiveTransforms()/cvTransform() (I've only tried this with the first function, but i guess it should olso work in the second one) and instead of providing a 3 channel array of source points as documented, use a 2 channel array with just the x and y coordinates.

shanaka wrote:

Vladimir Eremeev wrote:

It seems I am missing some basics, however, I haven't found any info on the net.

I have an affine transformation matrix which I am going to use to transform images with cvWarpAffine.
I have an image to warp, and I need to esimate the resulting image size in order to correctly allocate the memory.

My method is the following:
Estimate new coordinates for the corners and determine the surrounding rectangle.

I have created the matrix with the image corners' coordinates
/0 , 0\
|w, 0|
|0,  h|
\w, h/
where w and h are the width and height of the source image.

Then I was going to feed it to cvTransform, find a maximum and minimum values for each coordinate and estimate new width and height.

However, cvTransform was returning errors.

The failed C++ code is below

    IplImage *src=cvLoadImage("images/2/1.JPG");

// create initial matrix.
    CvMat *corners=cvCreateMat(4,2,CV_64FC1);
// fill it
    cvmSet(corners,0,0,0.);                      cvmSet(corners,0,1,0.);                              //(0,0)
    cvmSet(corners,1,0,double(src->width)); cvmSet(corners,1,1,0.);                              //(w,0)
    cvmSet(corners,2,0,0.);      cvmSet(corners,2,1,double(src->height));                      //(0,h)
    cvmSet(corners,3,0,double(src->width)); svmSet(corners,3,1,double(src->height));     //(w,h)

    CvMat *corners_tr=cvCreateMat(4,2,CV_64FC1);

// and calculate the transformation
    cvTransform(corners,corners_tr,res);

// then find min and max
    int minh=cvmGet(corners_tr,0,0);
    int maxh=minh,minw=minh,maxw=minh;

    for(i=0;i<4;i++){
    int c=cvmGet(corners_tr,0,i);
    if(maxw < c) maxw=c;
    if(minw > c) minw=c;
    c=cvmGet(corners_tr,1,i);
    if(maxh < c) maxh=c;
    if(minh > c) minh=c;
    }

// and calculate dst_size
    CvSize dst_size=cvSize(maxw-minw,maxh-minh);

I tried transposing the matrix 'corners' or adding a new row with 1s - no success.

I think ur roblom is that u r using 4x2 vector where as u should be using 4x1 3-channel vector with (x,y,1) cordinates :)

I ve used this function (for perspective transformation using a homography to map form one image to another) but for me it gives unexpected values. and when i used cvPerspectiveTransform() it all the transformed cordinates as 0,0.
i've tried to manually transform one point using cvGEMM() function acording to the equation p' = H x p and it works fine but when i give the data to this function it gives wrong cordinates..... is there any idea about it?