EE6850 Home | Next:System
Realize a basic content-based image retrival(CBIR) system, and have a better understanding of the problems solved or being solved in this area.
CBIR in a nutshell:
Task: Given an image(the query), find most similar images in the database.
Method: Extract a pre-selected set of features from each of the query image and images in the database, use some or all of the features to construct a feature space, and the closest points in this high-dimensional feature space are returned as the closest match.
Issues: how to select features, how to construct the space, how to measusre the distances in the space, how to switch between different set of features, different spaces, or different distance measures ... etc, etc.
Table of Contents:
System diagram and summary of algorithm.2. Features
Features: color histogram and edge histogram.
Distance measures: L-2 distance, cosine distance, histogram intersection.3. Decision-making
Issues on combining different distance measures.4. Results
Results of the benchmarking images.
5. Summary and Discussion
Lessons to take home.
A brief note on implementation, and source code.
A statement from Ghost : "for conciseness, optional example/results are included in seperate pages indicated by my presence, please click me to see them, enjoy :)"
About results: most of the result images are in PNG format, please let me know if your browser doesn't support PNG (Netscape 4.7+ or IE 5+ should work), thanks.