Foreground Background Segmentation using Graph Cuts (due in 1 week)
----------------------------------------------------------

The goal of this homework is for you to get familiar 
with the GraphCut algorithm used for optimal labelling. 

You will be asked to implement the foreground/background 
subtraction algorithm and run it on all images in the image 
directory. 

This is a two label problem which can be solved optimally with GraphCuts. 

1. You can assume that the problem is formulated on a pixel grid with 4
or 8 connected neighbours.  

2. You will need to implement the computation of the data term assuming 
that the label likelihood is computed  as -log(p(pixel/foreground)/p(pixel|bakground)).
You can decide on the choice of distribution for represing the likelihood distributions
(e.g. histograms, of Gaussian Mixeture are some options)

3. To construct the unary term you can initialize the foreground region you will use for 
estimation of the likelihood  by hand by clicking at two corners of a rectangle containing the object (use function ginput). 

4. Hand in the mathematical expression of the energy function. 

5. Visualize the foreground/backgroup likelihood map

6. Show the results if final segmentation on all images in {\tt image} directory with the 
i) data independent pairwise term and 
ii) contrast sensitive pairwise term. 

The required code to get started can be found in cs884/gc_segmentation/.
You can also download the all teh data and code from cs884/gc_segmentation.tar.
To see how to use the GraphCuts you can look at the example 
gc_example.m for color based segmentation on an image.