EE6850 Home | Previsous:Result |Up: Intro| Next:Appendix

CBIR #5: Summary

Summary of the experimental CBIR algorithm:
1. Features: HSV color histogram, edge histogram;
2. Distance measure used: histogram intersection;
3. Combine 2 features: mini-max distance among images among possible weights on [0,1];

Comments on general CBIR system:
1. It works, and often produces fairly good result;
2. Still a hard problem, the gap between low level features and perception-interpretation is not bridged;
3. Difficult to find features that correctly represent perception, and we have little idea about the geometry of feature space;
4. Difficult to find fair distance measures;
5. Yet to have a systematic way to combine(choose)different features or different models;
6. There are on-going researche activities to improve relevance feedback and other components;
7. It is also hard to compare the performance of different CBIR systems, mainly because it is hard to construct a global benchmark database and groundtruth.

Next Page: Appendix>>

HW3 Intro System Features Decision Results Summary Appendix
Last update: 10/29/2001 5:46 PM