Image compression using one-dimensional vector quantization
Date
1991
Authors
El-Maleh, Aiman
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Vector Quantization (VQ) is a well-known image coding technique. Previously, vector quantizers operated on two-dimensional (2-D) blocks spanning several adjacent lines in an image. In this work, we investigate the capability of one-dimensional (1-D) VQ as an image coding technique. 1-D VQ operates on blocks spanning a portion of a single scanning line. The advantages of 1-D VQ are considerable reduction in implementation complexity, reduced storage requirement and a faster encoder response. Problems encountered in 1-D VQ are discussed and compared to those encountered in 2-D VQ. 2-D VQ showed a better performance than 1-D VQ. The PSNR was on the average 1.6 dB higher in images coded using 2-D than in those coded using 1-D VQ.
In order to improve the perceptual quality of coded images using 1-D VQ, a special block accessing scheme is used. Furthermore, product structures are used to improve the performance of 1-D VQ and reduce the computational complexity. A novel product structure based on scaling (SVQ) or scaling and rotating (SRVQ) the vectors before quantization is introduced. Furthermore, the mean/shape vector quantizer (M/SVQ) is used and compared to our new structure. By using product structures in 1-D VQ, the signal-to-noise ratio (SNR) of coded images is increased by almost 1 dB using SVQ and by almost 2 dB using M/SVQ for the same coding complexity. From this work we show that, by using the b lock accessing scheme and product structures, 1-D VQ produces good quality coded images at compression ratios in the range of 8:1 (1 bpp).