README

This code illustrates the following paper:

	Making Background Subtraction Robust to Sudden Illumination Changes,
	Julien Pilet, Christoph Strecha and Pascal Fua, 
       	European Conference on Computer Vision, Marseille, France, October 2008.
	
The package contains two programs: emvisi2 and learn. "emvisi2" segments
foreground and background. "learn" takes a set of background/input/ground truth
triplets to estimate the behavior of NCC on your data. The resulting histograms
are saved in several files, including ncc_proba.cpp which emvisi2 uses.

A default ncc_proba.cpp is provided. Expect performance improvement if one
adapted to your own data.

You can optionally smooth the result with a graphcut. Have a look at the
Makefile.

If you have trouble compiling fwncc.cpp, try with the -msse2 compiler flag.

Please respect the GPL terms.

Good luck !
Julien Pilet. November 2008.