EE 589/689 Foundations of computer vision (Fall quarter 2006): project 2

Project 2 can be done in groups of up to two. By Wednesday Nov. 15, 2006, each group should either choose one of the following projects, or suggest me a different project:

• Image segmentation with spectral clustering
• Image segmentation with mean-shift clustering
• Image segmentation with hierarchical clustering
• Image segmentation with Gaussian mixtures and EM (see book exercise 16.5)
• Background subtraction with algorithm 14.1 and with EM (see book exercises 14.5 and 16.4)
• Fitting several lines with the Hough transform (see book exercises 15.9, 15.10)
• Line fitting with an M-estimator.
• Line fitting with RANSAC (see book exercise 16.9).
• Line fitting with EM (see book exercises 16.6-8).
• 2D Kalman filter to track blobs with background subtraction (see book exercises 17.5-6)
• Mark skin pixels with a classifier (see book exercise 22.6)
• Implement a face finder (see book exercise 22.7)

No matter what method you choose, evaluate its performance under different parameter values (things like the number of clusters/components, affinity function and width, thresholds, etc.), different feature spaces (intensity/colour/position...), different noise levels, etc. as they may apply, and also try a few different images.

Before starting to work on the project (and also as you progress with it), discuss it with me to make sure you submit something which is neither too little nor too much effort.

Once you have finished the project, email me:

• A compressed file (gzip or zip) containing your Matlab code (including a sample driver file illustrating how it works) and test images.
• A PDF file containing your evaluation & comments. Be comprehensive and probing in your experiments and evaluation, but concise in your description.

Deadline: Monday Dec. 11, 2006.

OHSU | OGI | Dept. of CSEE | Adaptive Systems Group | MACP's Home Page