Optimal feature-based vector quantization of image coding using integral projections

This paper presents a fast algorithm of vector quantization for image coding that relies on three conditions of early termination to confine the search space, resulting in acceleration of the encoding process. These termination conditions are derived based on the gray-level features that are extracted from the individual vectors of pixels. Each incoming vector is compared to the codebook entries first using these fast tests. The codebook entries that fail one or more of the fast tests can be rejected without further consideration. Thus, time-consuming computations of the squared Euclidean distance in the proposed system are performed on only a few codebook entries that first pass all three fast tests. Verified results for the system show over 97% reduction of execution time compared to the full search algorithm and 50% compared to the mean-ordered partial codebook search method in image VQ encoding.