Vector quantisation-based neuro-wavelet model with cumulative distribution function for efficient image compression

Abstract

An efficient image compression technique is required for storage and transmission of raw images that need enormous amounts of disk space. The compression algorithms for different types of images take a longer time to converge when they are compressed using radial basis function neural network (RBFNN) along with wavelet. The convergence of the network as well as the compression ratio can be improved, by estimating cumulative distribution function (CDF) for the image and CDF is used to map the image pixels. The main contribution of this paper is in developing a compression technique that combines the specific features of wavelet transform, RBFNN and vector quantisation using CDF. The distinct way in which the low and high frequency components are handled in this paper, makes it an efficient technique for compression. It is been demonstrated that the RBFNN, along with wavelet, not only yields better peak signal to noise ratio at high compression ratio but also reduces computation time when the mapped image pixels are used in relation to the unmapped image pixels.