Directional Image Coding on Wavelet Transform Domain

Abstract

Vector quantization (VQ) techniques have been widely studied by a lot of researchers to efficiently encode images. Transform-domain VQ utilizes the transformation of image to compact most of the energy of the signal into a few of almost-but-not-perfectly de-correlated coefficients and then VQ to exploit the remaining correlations to the utmost. A novel method of directional image coding on the wavelet transform domain is devised for efficient compression of image data. Directionally decomposed versions of an image are obtained by manipulating the wavelet transform coefficients, and then the coefficients of each version are segmented into vectors of reasonable dimension. Finally, the segmented vectors are quantized based on the gain-shape VQ. The proposed method yields excellent quality of reconstructed images at very low bit rates, much better than JPEG or other transform domain VQ algorithms.