An Improved Fast Encoding Algorithm for Vector Quantization Using 2-Pixel-Merging Sum Pyramid and Manhattan-Distance-First Check

Zhibin Pan, Koji Kotani, Tadahiro Ohmi

Research output: Contribution to journalArticlepeer-review

Abstract

Vector quantization (VQ) features a very heavy encoding process. In previous work [3], an efficient encoding algorithm using mean pyramid has been developed. To improve it further, a fast search algorithm is proposed in this letter. Specifically speaking, four major modifications are made. First, to rearrange the original codebook directly along the sorted real sums to reduce the search scope and then update the lower and upper bound dynamically. Second, to use sum instead of the mean that includes roundoff error to thoroughly avoid a possible mismatched winner. Third, to construct a sum pyramid using 2-pixel-merging other than 4-pixel-merging way to generate more in-between levels. Fourth, to introduce the Cauchy-Schwarz inequality to bridge Euclidean and Manhattan distance together so that the difference check between 2 vectors can be pre-conducted only by much lighter Manhattan distance computation. Experimental results show that the proposed algorithm is more search-efficient.

Original languageEnglish
Pages (from-to)494-499
Number of pages6
JournalIEICE Transactions on Information and Systems
VolumeE87-D
Issue number2
Publication statusPublished - 2004 Feb

Keywords

  • 2-pixel-merging
  • Fast encoding
  • Manhattan-distance-first
  • Sum pyramid
  • The Cauchy-Schwarz inequality
  • VQ

ASJC Scopus subject areas

  • Software
  • Hardware and Architecture
  • Computer Vision and Pattern Recognition
  • Electrical and Electronic Engineering
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'An Improved Fast Encoding Algorithm for Vector Quantization Using 2-Pixel-Merging Sum Pyramid and Manhattan-Distance-First Check'. Together they form a unique fingerprint.

Cite this