Variance oriented dynamic codebook adaptive vector quantization system for image coding

A simple and fast adaptive vector quantization system is presented for image coding. The proposed system is composed of three modules: a fast codebook generation, codebook component quantization and a dynamic codebook mechanism. In the codebook generation stage, a variance-based initialization technique is developed to improve the convergence rate and achieve a better local minimum, and a key-indexed searching method is also proposed to reduce the computations at each iteration for the generalized Lloyd algorithm. After codebook generation, each image vector is then quantized by a dynamic addressing approach with the concept of diminishing codebook size. Meanwhile, a perceptual coding scheme is exploited to encode efficaciously the codebook components for further reducing the bit-rate requirement of the system. Simulation results demonstrate that a gain of around 2 dB in PSNR upon the universal vector quantizer, together with a better quality especially in the high contrast areas, is acquired; while the general execution time of nearly 90% less than that of a full search codebook design, being comparable to that of a standard coding by the universal VQ, is obtained.