1

I am wanting to automatically align stereo images for best fit. The images are taken only a few inches apart and in the same lighting conditions etc but there might be slight rotations and shifts etc. Can someone describe the best approach for this sort of task? I assume one could use template matching. Would a brute force approach work just fine? Would it be better to detect features in each and try to align the features? I am not looking to rectify the images, just find the closest alignment. Any direction about methodology would be very appreciated.

Edit: Basically this will be a similar method to creating a panorama except that I do not need to warp the image.

Andrew King
  • 375
  • 2
  • 15
  • 1
    Since warping is not an option, I am guessing image registration using cross-correlation is the way to go. Just curious, how far these images are off with respect to each other ? as in is it in pixels or you want subpixel registration. – harshkn Apr 06 '17 at 03:55
  • @h4rshkn That looks like a great direction to go. No need for subpixel registration, they will be off by up to 15% horizontally and vertically and will also could be somewhat rotated. – Andrew King Apr 06 '17 at 04:12
  • 1
    Cross-correlation can be a good approach, provided you can handle the rotations. Keep in mind anyway that exact alignment *cannot* be achieved as depth creates parallax. –  Apr 06 '17 at 13:27

0 Answers0