Justin Jang (12/06/2001)

CS 4495/7495: Computer Vision
Final Project: Counting with Watershed Segmentation
Due: December 06, 2001
Notes: I ran my program with Matlab 5.3 in the Intel Lab at CoC.


Running time info:

The EDM took over a minute per image on average.
The Watershed took several minutes per image on average with the EDMs.
The Watershed took more than 5 minutes per image on average with the grayscale valleys.
All other computations took orders of magnitude less time.


Overview of Techniques:

Binary Morphology Method:
Convert to grayscale
Histogram equalization
Threshold to form binary image (the objects we're counting have the one values)
Binary Morphology (close then open, then invert)
Compute EDM from binary image (and invert w/ max minus vals, also normalize if necessary)
Watershed Segmentation
Count the segments

Grayscale Mophology Method:
Convert to grayscale
Grayscale Mophology (add tophat transform, subtract bothat transform)
Watershed Segmentation
Count the segments


See index.html for more information.