search

thunder-project / thunder

scalable analysis of images and time series
http://thunder-project.org
Apache License 2.0
814 stars 184 forks source link

Add sub-pixel registration method #205

Open andrewgiessel opened 9 years ago

andrewgiessel commented 9 years ago

I propose that we implement the following as a thunder registration method:

Manuel Guizar-Sicairos, Samuel T. Thurman, and James R. Fienup, "Efficient subpixel image registration algorithms," Opt. Lett. 33, 156-158 (2008). link

MATLAB implementation Original Python Port My generalization for neuro data in d_code

Essentially, images are transformed using an FFT, up-sampled to an arbitrary degree and aligned via cross correlation, the benefit of this algorithm is the efficiency of the up-sampling, and the fact that it is sub-pixel. It works quite well and porting the code should be straight forward.

freeman-lab commented 9 years ago

Thanks @andrewgiessel , I think this would be terrific to have! Take a look at the image registration API and maybe try putting together an implementation? Happy to jump into a discussion in the chat room if you get stuck. You can also check out this WIP PR from @poolio that adds another, different registration method (https://github.com/thunder-project/thunder/pull/100).

cc @broxtronix