Fast Encoding Method of Vector Quantization Based on Optimal Subvector Partition

The encoding process of vector quantization (VQ) is very expensive computationally. By using the statistical features of the sum and the variance of a k-D whole vector, IEENNS method has been proposed to reject unlikely codewords. To further enhance the performance of IEENNS method, by first partitioning a k-D whole vector in half to generate its two (k/2)-D fixed subvectors and then directly applying IEENNS method once again to each sub vector, a complete- version C-SIEENNS method and a simplified-version S- SIEENNS method have been proposed. By offline sorting the elements of a codeword before subvector partition, an adaptive A-SIEENNS method has been reported recently as well. However, there is no guarantee that these subvector partition methods are optimal. Thus, this paper proposes a practical criterion to optimally partition a k-D whole vector into a k₂-D first subvector and a k2-D second subvector (k₁+k₂=k) by maximizing the energy included in the two partial sums of a codeword. Experimental results confirmed that this work can improve search efficiency significantly compared to the latest A-STEENNS method.