Hector Martin

Seeing post about image stitching for IC's, has me thinking about an automated image stitching rig. Use lasers at points across the face to evaulate surface shape and create a more seemless automated stitch. Thoughts on how this would work?

I get the feeling this is a problem we can solve with (better) code. Planar stitching isn't very hard. Might want a calibration target to determine the lens distortion parameters first and then it should be easy.

thats what hugin attempts to do, but there are still a lot more variables than you'd think. Consistent movement in the x-y plane would help a lot, but you also have microwarps in the surface etc that can be serious issues for image stitching.

hugin tries to take 2 images that you tell it are adjacent, and use a set of key pairs of points to distort, deblur, etc using excess information. You can get somewhat decent results, but not better than what you can get with a hi-res scanner, which should absolutely be doable.

Honestly, hugin already does pretty good job. I did this one with hugin. Not the highest resolution, but it's also a very dense IC (Wii U GPU). Literally just pushing the die around with tweezers. https://marcan.st/transf/latte_stitched.jpg …

how many images did you stitch it out of? because when working with the switch PCB I ran in to several issues. required a huge amount of manual work to get something more complete. admittedly, my microscope was a piece of shit, lighting was terrible, etc etc.