Region Identification in CBIR Systems Using Pixel Descriptors

摘要

Content based image retrieval (CBIR) systems have attracted the attention of both commercial developers and academic research community. Texture and color are important cues for CBIR systems, but the contribution of color to the texture perception has been mostly ignored. We have introduced the notion of Pixel Descriptor(PID), which combines the human perception of color and texture into a single vector. Extracting individual image objects from an unconstrained image is difficult with the current state of the art. Such complete object identification is not necessary for the purposes of CBIR systems, and identification of homogeneous regions in the feature space is adequate. In this paper, we present an algorithm to extract regions from the natural images using PIDs. This technique is being implemented in the CHITRA CBIR system we are developing at RMIT.