image stitching for low feature images

asked 2019-04-19 18:15:23 -0500

I am trying to do image stitching of two images that have been shifted vertically and don't really contain a whole lot of features. Harris doesn't find anything, the blob detector finds random spots... I also tried PSNR and SSIM without success (the results are always very similar).

The images are pictures of walls. I can see where they should match by eye, so there is some detail in the pictures (grain and so on, different tones of the same color, etc).

How would you do it?

edit retag flag offensive close merge delete


you can find a tutorial here

LBerger gravatar imageLBerger ( 2019-04-20 02:13:21 -0500 )edit


I am also facing the same problem. Let's say I have 4 images that I want to stitch. [0 to 3] images[1-3] all contain enough information (features), but image [0] cannot be stitched with the rest and opencv stitcher is answering "need more images".

As Chourizo, i can guess by eye where it should join, so I think there should be a way to determine matching surfaces..?

@Chourizo, any update on your side?

gmz gravatar imagegmz ( 2019-05-24 08:40:36 -0500 )edit

@gmz , however, please do not post answers here, if you have a question or a comment, thank you.

berak gravatar imageberak ( 2019-05-24 08:57:05 -0500 )edit

@berak, ok I will mind next time ;)

gmz gravatar imagegmz ( 2019-05-24 08:59:38 -0500 )edit

@LBerger, your comment just points to the base stitching opencv code samples, but does not answers the question...

gmz gravatar imagegmz ( 2019-05-24 09:01:34 -0500 )edit