A Linear Time Side Match Vector Quantization Implementation

Abstract

Listen

Translate article

Side match vector quantization (SMVQ) is slower than vector quantization (VQ). For each SMVQ index, it generates a state codebook SCB. The N codewords from the main codebook D are sorted by their side match distortions SMDs. The average execution times with N = 512 and 1024 are 18.528 and 46.052 seconds. In this paper, we propose an approach that can exploit any integer sorting algorithm. First, it assigns a modified side match distortion integer MSMD to each codeword. A codeword’s MSMD is a function of its SMD. Next, it sorts the codewords in D by their MSMDs. This step can use any integer sorting algorithm. Our implementation uses the bucket sort algorithm. It reduces the SMVQ’s time complexity from O(Nlog2N) to O(N + K), where K is the number of buckets. Experimental results verify that our implementation is faster. The average execution times with N = 512 and 1024 are 4.279 and 7.247 seconds. That is double the time of VQ. Also, the average execution times with N = 2048 and 4096 are 19.833 and 51.710 seconds.