Subband Coding of Images Using Predictive Vector Quantization

Abstract

In this paper, two bfferent schemes for predictive vector quantization (VQ) of subband decomposed images are investigated. The aim is to reduce the quantization error by incorporating memory into the VQ scheme. The first scheme is a form of finite-state VQ FSVQ) which we will call subband FSVQ (SB-FSVQ) and the second is a form of predictive VQ (PVQ) applied to image subbands. We will refer to the second scheme as subband PVQ (SB-PVQ). It was found that both techniques outperform conventional subband VQ (memory-less) and spatial domain VQ in PSNR and perceptual tesms. It was also found that despite SB-FSVQ's ability to capture non-linear dependencies, SB-PVQ performs slightly better.