Lengthwise image stitching

New user here. I’m trying to use the Pairwise Stitching plugin to stitch together two sequential images to form an image with the same height but greater width, like this:

However, what I am getting looks like this this:

To be sure, the Y coordinates in the two images are not perfectly aligned. Even so, is it possible to get a stitched image with a small amount of vertical misalignment?

Thanks in advance,

AD

Incorrect stitching can be caused by too few common features in the overlapping region of two images. If you’re sure the images contain a substantial amount of overlapping signal, you can try playing with the plugin parameters (as explained on the wiki) to improve the result.

If there’s no or only little overlap, you can still create a montage (Image > Stacks > Images to Stack, followed by Image > Stacks > Make Montage…) or combine the images (Image > Stacks > Tools > Combine…).

Thanks for the reply. I already found one obvious problem in that the original images I was trying to stitch contained a legend with a bunch of SEM parameters (scale bar, etc) that is identical between the images, so I’m guessing the algorithm was detecting this legend as the strongest region of overlap, hence the problem. After cropping the images to get rid of the legend, the stitching algorithm does produce the panoramic stitch that I’m going for, but it still has a hard time identifying the correct stitch location. I tried playing with the “Check peak values” settings, and it seems that there’s a threshold value (8 to 9, in my case) where the stitch goes from basically panoramic more or less overlapping. I would have to agree that the overlap may simply be insufficient in my case; I will try to practice on images where I know the overlap is better.

Thanks!

3 Likes

I am not familiar with this particular plugin. However, one issue I have run into is that too many “common” features can also be an issue; depending on how the plugin does the registration. I found that sometimes I for better results operating on sub regions rather than the full images.

@Alexey_Dynkin,
how much overlap do you have in your images, is it ~10%?
You can also try TrakEM2 in Fiji plugins; it was giving nice results for us.

Thanks, I’ll give that a try. I think it was less than that in the previous images I was trying to stitch, more like 5% or less.

5% sounds a bit too low, it may work with some images, but to be safe it is better to use about 10%. I also had a similar issue during stitching of SB-EM datasets - there was a small overlap and spherical aberrations that made the stitching process very uncertain.