A Modified Fast Vector Quantization Algorithm Based on Nearest Partition Set Search

Abstract

In this paper we propose a modification to a fast vector quantization algorithm based on nearest partition set search. The fast algorithm searches the codebook to find the nearest set of codevectors for each codevector in the codebook. The nearest set of codevectors is called nearest set partition (NPS) which calculated each iteration. During each iteration the fast algorithm searches the NPS instead of searching the codebook which save training time. The NPS algorithm does well but with large codebook the saved timed consumed in calculating the NPS. So we proposed a modified algorithm to overcome this problem. The experimental results indicate that variation of NPS is slow with iteration. According to our results the calculation of NPS in each iteration is not necessary which save more training time without affecting the codebook quality.