next up previous
Next: About this document

CAP 6415 COMPUTER VISION (Spring 1997)
Programming Assignment 2 (Due on March 25, 1997)

In this assignment you will experiment with texture analysis, specifically the problem of texture classification. You will consider the ability of co-occurrence matrix based measures to discriminate between various natural textures.

A database of texture images is provided from the class home page at sarkar/cap6415.html

Each image is 64x64 and is in PGM format. There are images of four different textures, namely, Asphalt, Concrete, Grass, and Rock. There 12 images of each type of texture taken under different lighting conditions.

You may code in any language of your choice. I would suggest that you consider MATLAB or the Motif based code used in your image processing class (look into /home/cs/faculty/goldgof/Motif) as your building block. The Motif based code will allow you to display images, to interact with them, and to add your own procedures. A document about the Motif code is available at Procopy (maybe filed under Digital Image Processing or Dr. Goldgof).

As a part of this assignment you have to

  1. Write code to compute the co-occurrence matrix given an image and a displacement vector d. Also, write code to compute four texture measures from this co-occurrence matrix: energy, contrast, homogeneity, and entropy.

  2. Consider the following values for the vector d: (0, 1), (1, 0), (3, 0), (0, 3), and (2, 2). For each d, compute the four texture measures for each images. For each d, plot the four measures for all the images as a set of 2D or a 3D scatter plot. Note that you will need six 2D scatter plots or four 3D scatter plots to cover all projection of the 4D texture measure space.

  3. Write a report of at most 10 pages describing your experiments with the image set. Your report should be written professionally in a word processor. The discussion should be broken up into sections and all tables and figures should be discussed. In your writeup investigate the ability of the four texture measures to discriminate between the four texture classes. You should consider questions such as: Is there one measure which works best or does a combination of measures perform better?

You need to submit a written report and a hard copy of the code you have written. And you have to demonstrate your code with Matt Clark. The written report is due at 11 am on March 25, 1997. The demonstrations have to be done during Matt Clark's office hours (12:15 pm to 1 pm) on March 25 and March 26, 1997.

next up previous
Next: About this document

Sudeep Sarkar
Tue Feb 25 18:10:01 EST 1997