Applying Specific Region Frequency and Texture Features on Content-based Image Retrieval

In this paper, a specific region called affine noisyinvariant region is extracted from a query and database imagesto help accurate retrieval on different attacks. Then, only a 64×1codebook based feature vector is obtained from this specificregion applying vector quantization and codebook generationbased on the Linde-Buzo-Gray algorithm, which reducesretrieval feature comparison calculations. Also a number oftexture and frequency domain based features are computed andestablished for the region. Finally combination of these twogroups of feature vectors improves the retrieval systemefficiency. Besides, in order to optimize weighting combinationcoefficients of the feature vectors, the particle swarmoptimization algorithm is applied. The experimental results showa real-time content-based image retrieval system with higheraccuracy and acceptable retrieval time

Keywords:

content-based image retrieval , affine and noise content-based image retrieval , affine and noise feature , particle swarm optimization