Incremental codebook generation for vector quantization in large scale content based image retrieval

In this paper, we explain a novel technique for using vector quantization (VQ) for large scale content based information retrieval of any type of images, not restricted to the compressed domain. The main problem with VQ is the size of the source vectors that is used to generate the global codebook, which represents all images in the database. We have proposed a technique, where the local codebooks are generated and the locally global codebook for the images is computed from all the local codebooks. It gives better image quality than the global codebook. This incremental codebook generation process makes the index scalable, as new image codebooks can be used to generate the locally global codebook easily.