Image Coding Based On Directional Filter Banks And Wavelet Transforms

The technique of multi-media makes people's lives much creative, however, in the face of huge information of images, how to represent, store and transmit it more effectively? Image compression is very to solve this problem.Wavelet transforms have been applied and studied in the image coding field extensively because of its good feature of space-frequency and human visual system. Wavelet transforms of an image is to decompose the image into low frequency components and high frequency components, the former ones represent approximation of the image and the latter ones indicate details, and the energy of the image mainly lies in the low frequency components. Then, by making quantization and entropy coding on these wavelet coefficients, the work of compressing the image is finished. Contrast to wavelet transforms, Directional Filter Banks(DFB) can extract directional and edge information of images better. In image compression, wavelet transforms and DFB are both regarded as methods of image transformation.This thesis investigates image coding based on wavelet transform and directional filter banks, among various of directional filter banks, we adopt quincunx directional filter banks. The work can be mainly divided into two aspects: lattice vector quantization in vector quantization coding for images and the lifting realization of the quincunx directional filter banks.In image compression, lattice vector quantization (LVQ) reduces coding and computational complexity because of its regular structure in image compression. An important issue of LVQ is the tuning of its key parameter, the scaling factor, having the effect of altering the size of the basic quantization cell and influencing bit-rates and distortion while encoding. In this thesis, we propose a new method for tuning this key parameter, which is related to the cumulative probability density function of the radius of the codebook shell. This scheme is then applied to a LVQ scheme based on the use of a vector dead zone (DZLVQ) and results in an improved rate-distortion model. Experimental results illustrate this model is superior to the rate-distortion model in pyramidal dead zone lattice vector quantization.This thesis proposes a simple scheme with symmetric coefficients for the prediction and update operators in quincunx lifting, which is then applied into image coding based on Embedded Zerotree Wavelet (EZW) algorithm. In addition, we present a new quantization method for taking quantization threshold by combining the power of 2 with the square of integer. The former is used for image coding at higher bit rates and the latter at lower bit rates. The experimental results demonstrate that the proposed method can improve the quality of the reconstructed image.