An adaptive vector quantization technique with a fuzzy distortion measure for efficient image coding

Date

1996-08

Journal Title

Journal ISSN

Volume Title

Publisher

Texas Tech University

Abstract

Digital image compression techniques are currently experiencing significant growth due to diverse applications demanding efficient storage and transmission of increasing image data contents. These compression techniques involve representation of an image with reduced number of bits pdr pixel by exploiting the redundancy in data present within an image. In lossy^ compression, information theory predicts that the performance of vector quantization (VQ) is superior to that obtained using scalar quantization (SQ) in optimizing the rate distortion fimction. In practice, however, the existing VQ algorithms suffer from a number of serious problems, e.g., long search process, codebook initialization and getting trapped in local minima. This research develops an adaptive vector quantization technique for generating an optimal codebook by employing a neurofuzzy clustering approach to ensure minimum distortion.

In addition, a multiresolution decomposition of an image is used as a preprocessing stage for transforming the image into a form that is more suitable for quantization, coding and progressive transmission. The multiresolution wavelet decomposition of an image is performed using Daubechies coefficients prior to vector quantization and a multiresolution codebook scheme is used for quantizing the sub-images at different resolutions. This integrated approach of adaptive vector quantization with wavelet based pyramid image decomposition aids in significant facilitation of the compression and coding processes thereby allowing higher compression ratios with acceptable visual quality.

Experimental results of this new approach show significant improvement in performance as compared to a variable block size vector quantization (VBQ). The superior performance of this integrated algorithm has been validated by applying it to several classes of images and comparing the performance in terms of mean-squared error (MSB), peak-signal-to-noise-ratio (PSNR), bit rates and visual fidelity.

Description

Keywords

Image processing, Fuzzy systems, Computer algorithms, Image compression

Citation