pyimreg / python-register

Linear and non-linear image registration methods, using scipy and numpy.
Other
62 stars 34 forks source link

Explore optical flow for image registration #6

Closed nfaggian closed 12 years ago

nfaggian commented 13 years ago

Consider optical the optical flow between two images as the non-linear deformation - it can be implemented as a non-linear registration algorithm.

This paper looks good:

High Accuracy Optical Flow Estimation Based on a Theory for Warping Thomas Brox, Andrés Bruhn, Nils Papenberg, Joachim Weickert (2004) Computer VisionECCV 2004 3024 p. 25-36

stefanv commented 13 years ago

Two quick ways of getting this done:

1) Extend the scikits-image OpenCV wrapper 2) Use the OpenCV python wrapper directly (don't think this option handles numpy arrays)

http://opencv.willowgarage.com/documentation/c/video_motion_analysis_and_object_tracking.html

nfaggian commented 13 years ago

Sounds good to me - have to look at how this tracks with the paper, but certainly would be a good start.

If we implemented this I would like an interface (layer of indirection to opencv), which can point to the opencv implementation if it is available (installed) otherwise point to a pure numpy version.

nfaggian commented 12 years ago

Not sure if this is something we need to do anymore - should probably just focus on getting the cubic spline model working as well as possible.