Content-based image retrieval using region segmentation has been an active research area in the past few years. Constrasting to traditional approaches, which compute only global features of images, the region-based methods extract features of the segmented regions and perform similarity comparisons at the granularity of region. In this paper, we propose a novel region-based retrieval method, Self-Learned Region Importance (SLRI). In this method, image similarity measure is based on the region importance learned from users' feedback. The region importance that coincides that human perception con not only be used in a query session, but also be memorized and cumulated for future queries. Experimental results on a database of about 8,600 general-purposed images show the effectiveness of our method using relevance feedback.
展开▼
