A regionalized content-based image retrieval framework

Utilizing regionalized features in Content-based Image Retrieval (CBIR) has been a dynamic research area over the past years. Several systems have been developed using their specific segmentation and feature extraction methods. In this paper, a strategy to model a regionalized CBIR framework is presented. Here, segmentation and local feature extraction are not specified and considered as “blackboxes”, which allows application of any segmentation method and visual descriptors. The proposed framework further adopts a grouping approach in order to “correct” possible over segmentation faults and a spatial feature called region proximity to describe regions topology in a visual scenery by a block-based approach. Using the MUVIS framework the proposed approach is developed and tested as feature extraction module, and its retrieval performance is compared against two frame-based color-texture descriptors. Experiments are carried out on synthetic and natural image databases and results indicate that a promising retrieval performance can be obtained if the segmentation quality is reasonable; however texture descriptors in general are degraded whenever applied on arbitrary-shape regions.