Let’s see, how about “when the photos won’t stitch, there must be a hitch”? No? Well, anyway, the Photomerge ^TM option in Adobe Photoshop Elements serves to stitch multiple images together into one. It usually works remarkably well, but sometimes it just doesn’t do the job. In those cases, the problem can often be solved by perspectively correcting the individual images before merging them.

Here is a sequence of images of melons for sale in a supermarket, to be specific, in August at the Save-Mart in Angel’s Camp, California (map).

We are not talking art museum material here. I wanted the pictures for a food blog post on the varieties of melons available, and here we have Crenshaw, Orange-flesh, Casaba, Juan Canary, and Gailia melons in resplendent array. To take the picture without stitching I would have needed either a panoramic camera or a step ladder, neither of which I had with me.

The images are brought into Photoshop and automatically stitched into a panorama, per our stitching tutorial. The result is that the melon part of the image stitches reasonably well, but the signs are a mess:

Trying the perspective and cylindrical modes of stitching does no better. The problem is that the perspective distortion in each image is too great to make the match. The cure is to perspectively correct each image individually before attempting the merge. We have a tutorial with a video that shows how to use the Correct Camera Distortion filter in Photoshop. In this case, the vertical perspective correction of +75 made the verticals of the center image parallel. Exactly the same +75 correction was applied to the left and right images, yielding:

If the corrections are not identical in each image, the resulting images will be of different heights, and that is fatal to successful stitching. If +75 is used for one image, it must be used for all of them. We must crop out the “wings” on all the corrected images except the two outer edges. Then the automatic merge works, yielding:

Note that I cheated and filled in the upper right corner of the image using the Clone Tool. Another possibility is to take the image back to the Correct Camera Distortion filter and apply a -75 correction. That widens the top of the image.

Why doesn’t the merge always work? It is a limitation of present software. A way to solve the problem is to rewrite the software convert each image to a spherical projection, perform the merge, then project back to a tangent plane or a tangent cylinder. Everybody who understands what I just said should pass the word along to Adobe.