Significance-Preserving Codebook Using Generic Algorithm

Vector quantization (VQ) is an effective mechanism for image data compression. Each image block is replaced with a quantized index to compress the original information; individuals can quickly restore the condensed data to the host ones using this index table. The main merits of VQ lie in fixed compression ratio and fast decompression. The codebook generation is the principal component in a VQ mechanism. A codebook must consist of a set of representative codewords. Hence, how to construct a significant codebook is the primary challenge in designing a VQ mechanism. We propose a novel codebook generation method by adopting the generic algorithm (GA) to solve optimization problems. Each codeword is considered as a gene, while the codebook is regarded as an individual in the new mechanism. Furthermore, we improve the convergence speed of GA while training a codebook. According to experimental results, the new scheme can provide high- quality codebooks for vector quantization, compared to the well-known LBG algorithm and Ying et al.´s method.