Written by Mohamed Eldawy (eldawy@cs.wisc.edu) on 12-30-2006. Please email me with any questions.


What does this do?
------------------
This program does the following..
1. Read a list of *.ppm images
2. Applies a segmentation algorithm on each image
3. Transform images in L*U*V color space
4. Build a histogram of L*U*V values in each segment
5. Construct a feature vector for each segment, and output those in a file


What is its role?
-----------------
This is step #2 for finding the best image for a keyword. Here we build the feature vectors for each image. We will select the best vector in step 3

How to use it?
--------------
Once you have built the program, you can run it in the following way...
 
  segment sigma k min N

where

 sigma, k, min are parameters to the segmentation algorithm. (typically sigma = 0.5, k = 500, min = 20)
 N is number of images (this one depends on the result of the first step)

The program output a text file called "histogram.txt". This file contains all feature vectors. This will be used in step 3.

Issues
------
Remember that this program need *.ppm files (not *.jpg). The files output by step 1 need to be converted to *.ppm files. This can be done using the command line.
  
  convert img.jpg img.ppm

Credits
-------
This implementation of the segmentation algorithm itself was written by Daniel Huttenlocher and is available at http://www.cs.cornell.edu/~dph/

The details of the segmentation algoirthm can be found in this paper...
"Efficient Graph-based Image Segmentation"
http://www.cs.cornell.edu/~dph/papers/seg-ijcv.pdf