الفهرس 
الغلافOnly 14 pages are availabe for public view
 |  | from | 103 |  |  |
| | | from | 103 | | |
Abstract
ABSTRACT
This thesis presents integration of Region-based image retrieval (RBIR) with relevance feedback (RF) to enhance the performance of Content based image retrieve (CBIR). First, an image is segmented into non overlapping regions. The segmentation procedure is based on a hierarchical watershed algorithm that extracts regions automatically. Then, extracting features from each region. Two types of features are extracted that are visual features and importance of each region. Regions are matched by comparing the visual features like, color, texture and shape and the Integrated Region Matching (lRM) distance measure between the whole images is calculated. The relative importance of the above features is spatial location and size. Relevance feedback asks users to mark relevant and/or irrelevant retrieved images after each round of image retrieval. The system updates the feature weight in two scenarios. The first one, As long the user session is active and doesn’t expired yet, it update the weight with high learning factor. The second one, whenever the user has done the current search thread, the system updates the weight feature with a small learning factor in the image feature database. Information collected from the returned images is utilized to update weight of each feature to improve retrieval performance in the next round.