Codebook Generation for Vector Quantization by Sorting the Sum of Sub Vectors

Vector Quantization is a lossy image compression technique. In this paper, we propose a novel idea for generating a codebook by sorting the sum of sub vectors. The training vectors are sub divided into four sub vectors each consisting of four elements.The sum of sub vectors 2 and 4 are subtracted from the sum of sub vectors 1 and 2. The resultant values are used to sort the training vectors. From the sorted list, the training vectors at every nth position are selected to form the codebook. The experimental results and the comparisons show that this method gives better performance with respect to the time taken to generate the codebook and the PSNR value (quality of the reconstructed images). The computational complexity involved is also very less. The codebook generated using the proposed method is optimized using the iterative clustering method. The quality of the reconstructed image is improved to a significant value.