Optimized Feature Integration and Minimized Search Space in Content Based Image Retrieval

Abstract

Abstract The semantic gap between the user request and retrieval result is an important but unsolved problem in the content-based image retrieval (CBIR) systems. This paper introduces a new multi-level structure in a CBIR system to bridge the semantic gap using the combination of low-level visual contents of an image. The initial stage of the proposed system depends on the statistical information of the color images which gives the most prominent images for the further level of the process. In the next step, low-level features such as color and texture details are extracted using dominant color descriptor (DCD) and radial mean local binary pattern over the query and selected images. Subsequently, Particle Swarm Optimization (PSO) is applied over both the color and texture similarity measure between the query and selected images. Finally, this multi-level system is experimented on OT-scene and Corel-10k databases to assess the performance and it gives 78.43% and 52.34% average precision rate.