2

I would like to have an estimate of how fast it is to run an image registration algorithm on two HD images in Python. The registration needs to be very robust, too. Ideally running within under a second. I found a few library and did some tests (such as using cv2's first code snippet in this page), however results were disappointing.

For example, it did not work on the two following images, trying to fit the second one on the first one:

Would anyone know a fast algorithm that is robust, callable from Python? What about speedups related to image resolution? Ideally I would use HD images. I also found this, but did not got the time to test it: http://pyimreg.github.io/

Guillaume Chevalier
  • 9,613
  • 8
  • 51
  • 79
  • 1. Create a pyramid of resolutions and first find the homography in the smallest resolution, then use that homography as your initial guess for the next pyramid, and you keep going up until you're at full resolution. 2. Calculate homographies from features (i.e. sparse image registration as opposed to dense, which uses all pixels). See this great SO answer [here](https://stackoverflow.com/a/10314876/5087436) and some helpful OpenCV docs [here](http://docs.opencv.org/3.0-beta/modules/video/doc/motion_analysis_and_object_tracking.html) – alkasm Aug 31 '17 at 01:51
  • Just asking: your example should work with *rigid* image registration, right ? – Gabriel Devillers Aug 31 '17 at 08:00

0 Answers0