Relevance feedback (RF) and region-based image retrieval (RBIR) are two widely used methods to enhance the performance of content-based image retrieval (CBIR) systems. In this paper, these two methods are combined to have the promising result of CBIR. Rather than using a single positive feedback group, the proposed approach embeds RF in the RBIR scheme using multiple positive and negative groups. To guide users in grouping the positive feedbacks, the proposed system provides an objectively heuristic pre-clustering result automatically. Referring to these guiding clusters, the users can then easily and subjectively re-group the positive feedbacks in accordance with his/her particular interests. A region-weighting scheme reflecting the process of human visual perception is proposed to enhance the weighting importance assigned to the region whose pixels are closer to the attention center. Finally, a modified Group Biased Discriminant Analysis (GBDA) is developed and applied to the similarity measure between images constructed on the basis of the region-based relevance feedbacks.
All Science Journal Classification (ASJC) codes
- Computer Science Applications
- Artificial Intelligence