Project 7:  Segmentation by Clustering

 

        

 

Description

         Implement image segmentation using Normalized Cut, described in “Normalized Cuts & Image Segmentation” by J. Shi and J. Malik. Use intensity, color, and texture for similarity.

         Implement K-means using the same tokens.

         Test both algorithms on the standard collection of images from Berkley.

 Results

I don’t require comparing your results to the ground truth provided with the images, because such comparison is non-trivial    process. However, if you are considering doing your research in segmentation, I suggest implementing this evaluation (and you’ll get a large bonus). The details on the ground truth and evaluation process could be downloaded from here.

Data set

The Berkeley Segmentation Dataset and Benchmark

 

Useful Links

Class lecture on segmentation

Book chapter on segmentation

Status

Sivan and Sergei