Object Recognition

Problem
Object recognition is a basic problem in computer vision which has received a lot of attention and yet remains unsolved. Here is one (simplified) version of the problem: Given a set of objects that have been seen before, correctly identify their presence or absence in a novel input image. In this project you will experiment with some simple object recognition algorithms using the COIL-100 image database from Shree Nayar's group at Columbia. 

Here's what you need to do:

1. Read selections from the following two books
   • M. Tarr and H. Bülthoff (editors), Object Recognition in Man, Monkey, and Machine. MIT Press, 1999.
   • S. Ullman, High Level Vision, MIT Press, 2000.
2. Write an object recognition program for the COIL-100 dataset. Your program should accept half of the COIL images for training purposes. Given any image from the testing set, your program should output a label identifying the object. Possible representations of object appearance include histograms, principle components, etc. You are encouraged to examine the data and explore different feature choices. You should select three representative images for evaluation purposes, and I will give you 2-3 more. Your report should describe the performance of your system on these examples.
3. Experiment with texton approaches to object recognition (time permitting).

Evaluation

You will write a three page report describing your experiments. It should address the following issues:

• Describe at least three factors that make object recognition difficult.
• How good are people and animals at object recognition?
• What were the strengths and weaknesses of the object recognition approach you implemented?
• What is the best line of attack for further work on this problem?