Integrating Color Vector Quantization and Curvelet Transform for Image Retrieval

Since most of image databases today are pooly indexed or annotated, there is a great need for developing automated, content-based image retrieval (CBIR) systems to help users to get images they want. The focus of our research is mining image features which can represent image in the way of human visual perception. Our image retrieval approach depends on the extracted color and shape features. Vector Quantization (VQ) can provide a way of better exploiting the spatial information to generate different color histograms than scalar color quantization, thus VQ is employed in this work to extract color pattern of images. The shape feature of images is extracted by curvelet transform, as it has been proved that the curvelet transform is an almost optimal sparse representation of objects with edges. The extracted color and shape features are combined and weighted by using Genetic Algorithm (GA), then used for image similarity measurement. Experimental results show that the GA combined features can bring about good retrieval precision and speed simultaneously.