Content-Based Image Retrieval Based on Affine Noisy Invariant Color Region

The content-based image retrieval methods retrieve images by image features. In this paper, color and gray affine noisy invariant regions are extracted from a query and database images to help accurate retrieval on different attacks. Also a number of color statistical properties of the color region are computed and color feature vectors establishes for this region. Besides, for the gray region, only a 64  1 feature vector is obtained applying vector quantization and codebook generation based on the LindeBuzo-Gray algorithm, which reduces retrieval feature comparison calculations. Finally combination of both gray and color affine noisy invariant regions improves the retrieval system efficiency. In order to optimize weighting combination coefficients of the color feature vectors with the gray feature vector, the particle swarm optimization algorithm is applied. The experimental results show a real-time content-based image retrieval system with higher accuracy and acceptable retrieval time.