How to reduce camera distortion while stitching images?

I want to create panorama using this matlab code Image mosaicing but it leaves a camera distortion (cycled red) How can I improve it (maybe there is matlab code for that transformation?) and in which step (before homography or after?).

I would be very grateful for help, really need this!

Answer 1

[Image Analyst]

There is nothing in MATLAB, that I'm aware of, that fixes that. You're going to have to explore the literature to find a published algorithm that stitches better than your algorithm: http://iris.usc.edu/Vision-Notes/bibliography/contents.html

[Julia]

Really, no code? I thought there is just missing part like affine transformation or smth (can't figure it out what kind of transformation should be used to reduce perspective distortion on left side of image) in that code.

[Image Analyst]

Sure, there's imtransform() and things like that, but it won't fix problems like what you showed: stitching misalignment. It's better to fix it in the original stitching process than to try to fix in a patchwork manner after the stitched image has already been created.

[Julia]

Yes, I want to stitch it in stiching process. No need to modify already created image, just want to make some transforms while stitching 1st image with 2nd, then 1_2 with 3, and so on.. Just don't understand what transform I should apply before stitching 2 images tohether to make result more accurate :) Now in the code it calculates only homography

[Julia]

Now the code is like that:

[matchLoc1 matchLoc2] = siftMatch(img1, img2);
 ....
[H corrPtIdx] = findHomography(matchLoc2',matchLoc1');
tform = maketform('projective',H');
img21 = imtransform(img2,tform);

Homography matrix looks like that:

H =

    1.2125   -0.0352 -125.0579
    0.1028    1.1287  -25.0789
    0.0006   -0.0000    1.0000

[Image Analyst]

I haven't studied up on that. I don't know what your File Exchange code does (and don't have time to figure it out) and I don't know how the published algorithms work. I'm sure it's a variety of complicated methods. Good luck though.

[Dakota Warner]

Did you solved this problem@Julia, @Image Analyst? i am interested in this solution