Zhibin PAN^*, Koji KOTANI¹ and Tadahiro OHMI

New Industry Creation Hatchery Center, Tohoku University, Aza-aoba 10, Aramaki, Aoba-ku, Sendai 980-8579, Japan
¹Department of Electronic Engineering, Graduate School of Engineering, Tohoku University, Aza-aoba 10, Aramaki, Aoba-ku, Sendai 980-8579, Japan

(Received August 14, 2003; Accepted March 23, 2004; Revised March 19, 2004)

Based on the concept of multi-resolutions, full search (FS) equivalent fast search methods for vector quantization (VQ) using mean-type pyramid have been developed in previous work.¹⁾ In this paper, a modified memory-efficient sum pyramid is proposed, which is exact for computing the difference in integer form and features one more new resolution L₀₁ in-between the top 2 levels L₀ and L₁ of a conventional pyramid. Benefiting from a recursive difference computation at the top 3 levels, the proposed method can reduce the total computational cost obviously. Furthermore, a memory-efficient way of storage for the top 3 levels is integrated to reduce extra memory requirement.

Key words: fast encoding, vector quantization, modified memory-efficient sum pyramid

^*Corresponding author, E-mail address: pzb@fff.niche.tohoku.ac.jp