Notes

All of the projects proposed below are related to computer vision and pattern recognition. However, they also include important components related to algorithm design, database design, artificial intelligence, machine learning, data mining, software engineering, graph theory, user interfaces, mathematics, statistics, etc.
I am not expecting you to have a computer vision or pattern recognition background but you should be motivated and enthusiastic about learning new topics.
Specifically, these projects require strong background in C++ programming and interest in computer vision, pattern recognition, machine learning, database design and user interfaces.
I expect the students to work consistently throughout two semesters and to have periodic meetings with me to discuss their progress.
These projects can be extended to M.S./Ph.D. research in the future.
Please contact me if you are interested in any of the projects below or would like to propose your own project.

Projects

Multimedia classification and retrieval
We are participating in the TREC Video Retrieval Evaluation Contest organized by the U.S. National Institute of Standards and Technology. Requirements include semantic classification and retrieval in very large video databases. (For example, one video of a 30 minute news broadcast is about 400 MB. Imagine searching in a database of 300 such videos.) Current techniques either require manual labeling of database content or perform searches using only low-level information such as color. New techniques and efficient tools are required to perform semantic classification and interactive searching in large archives.
Image classification and retrieval
We are interested in performing content-based classification and retrieval in large archives of satellite and medical images. (For example, nearly 3 Terabytes of data are being sent to Earth by NASA's satellites every day.) Current systems allow queries only on metadata such as geographical coordinates, time of acquisition, sensor type, etc. (Have you tried Google Earth?) New techniques and efficient tools are required to perform automatic semantic classification and retrieval based on image content (such as finding satellite scenes with similar land development, analyzing effects of a natural disaster, finding patients with similar cell structure for cancer studies).

TBD

Please check the list of important deadlines for the 2006-2007 Academic Year.