Analyzing Spatially-varying Blur: Source Code

Back to Project Page
Note: All code on this page is (as of now) released only for non-commercial research use. Please contact ayanc[at]eecs[dot]harvard[dot]edu if you need to use the code for commercial purposes. If you use this code to generate any results included in an academic publication, please cite the following paper:
Ayan Chakrabarti, Todd Zickler, and William T. Freeman, "Analyzing Spatially-varying Blur," in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 2010.

Blur Cue Code: For 1-D windows and kernels


This code implements the local blur likelihood measure described in the paper. Given an image and a candidate kernel, it computes the likelihood for every local window in the image being blurred by that kernel. This implementation considers 1-D windows and kernels (at horizontal, vertical or 45/135 degree orientations). Please see the README and demo.m files for details.


Segmentation Code:


This code implements the algorithm combining the blur cue above with color information, to carry out the full segmentation. Includes the blur cue code above, as well as a graph-cut implementation using the library from Vladimir Kolmogorov. Please see the README file for details on compiling the graph-cut code, and demo.m for sample usage.