Image Retrieval with Image Attention Based on User Input in a Multiresolution Grid Framework

Region-based image retrieval (RBIR) is widely used for enhancing the performance of content based image retrieval (CBIR). In this paper, we present a novel scheme to enhance retrieval performance using user specified image attention within a multiresolution grid framework. The coarse segmentation by the grid, yielding image tiles, will serve as basic building blocks for feature computation at different resolutions. The user then specifies the tiles containing concepts for search. The image is partitioned into subimages based on this image attention information. An integrated subimage matching scheme is used for image similarity. The experimental results demonstrate the efficacy of the method.