EE/CSE 576: Image Understanding

Spring 2001


EE/CSE 576: Final Project

During the second half of the quarter you will do a 4-5 week class project, to allow you to do something with more depth than the small assignments. Projects are to be defined by you, proposed to me, and approved before starting. You can start any time, but the official due date for proposals is May 4. You can choose something that comes from your own interests, or there are default projects such as segmentation of images according to color and texture properties. (See below.)

The idea of the project is to design and implement a computer vision system that can do the task you have chosen. You may use packages for utilities whenever possible, but there should be some programming of your own in the project, too. You should not just get it working on one image, but apply it to a small set of test images and evaluate the results.

The final project is due on June 6 at 5pm, by which time you should have turned in a typed project report of 5-10 pages. Project reports should contain at least the following:

  1. Introduction explaining the problem you tried to solve
  2. Discussion of any relevant literature you have read
  3. Description of the techniques you used to solve the problem
  4. Experiments and Results
  5. Conclusions and Future Work
  6. Appendix with the code you developed

Joint Projects

Projects may be done independently or in pairs. If two people work on a project, their separate parts should be easily identifiable and documented in the final report. For example, one person could do segmentation and the other use it for classification. If one partner fails, the other still has something, since he could select regions by hand for his classification part.

Default Projects

  1. Segmentation by color and texture
    Design and implement a segmentation system that uses both color and texture to segment color images into regions. A region is an area of the image that is deemed homogeneous in both color and texture according to the method used. You can implement a method from the literature (or more than one to compare if they are not very difficult) or can you choose to develop your own method. We have lots of color images including images from previous quarters, my football sequences, a set of paintings from Italy, and a set of gem and other exhibit images from the Smithsonian and my groundtruth database. A two-person project could extend this to an image retrieval system, like Blobworld.
  2. Content-based retrieval of images with "object regions"
    The goal of this project is to develop a classifier that can find regions of an image (without full segmentation) of a particular color/texture that represents some known object. The known objects should be things like tiger, zebra, tulips, etc. that can be represented well by color/texture. The classifier should be learned from training data.
  3. Specific-object recognizers
    For a given interesting class of objects, i.e. cars, faces, clothed people, animals, develop a sophisticated (i.e. not just color and texture) recognition algorithm and test it thoroughly showing both its successes and failures.

Examples of Past Student Projects


Last updated on .