Neuro-Wavelet based vector quantizer design for image compression

Abstract

This paper presents a novel idea of designing codebook, which is heart of vector quantization scheme based on Kohonens self organizing feature maps (SOFM) and wavelet transform. Being named as neuro wavelet generic codebook for compression of gray images, it can also be extended to the compression of color images and video frames. The code vectors are generated by evaluating the characteristics of the specific image sub samples, which are determined through rigorous mathematical operations and training the selected samples by Kohonen's SOFM artificial neural network with adjustable learning rate and initializations conditions followed by application of discrete wavelet transform. The testing of the codebook is done with variety of images and the compression performance is evaluated by using objective and subjective quality measures such as image fidelity, structural content, mean structural similarity index, universal quality index, spatial frequency measure and spectral activity measure along with PSNR. The proposed design is such that it can be implemented with less complexity and cost using VLSI techniques.