Synthesis and development of image coding algorithms based on 2D discrete wavelet transform.

摘要

Wavelet transforms have made significant contributions to the development of image compression algorithms. The key advantages of wavelet-based image compression are its multi-resolution analysis and good approximations to the original image that can be compactly represented. An entropy codec can be beneficial to the wavelet image representation to achieve an optimal codeword assignment for the compression of images. In this thesis, two wavelet image-coding techniques will be presented. They are the wavelet sub-band energy feature remapping (EFRM) and the wavelet maxima mapping quantisation (WMMQ) algorithms. The EFRM algorithm is a pre-processing technique for an adaptive arithmetic entropy codec. The objective of the EFRM algorithm is to remap the energy features of the wavelet image sub-bands into a coherence data structure that reduces the sub-band orientation sensitivity. As a result, the coding gain of an adaptive arithmetic entropy codec can be improved. The WMMQ algorithm is an improvement of the scalar-quantisation strategy for the wavelet image low-pass sub-band. The objective of the WMMQ algorithm is to selectively refine the amplitude of quantised DC coefficients in a wavelet transformed image low-pass sub-band. Consequently, the image background regularity of a compressed-reconstructed image can be significantly preserved with higher a DC level in the image low-pass sub-band. Empirical analyses with various experiments on standard test images have been conducted to evaluate the performance of the two proposed techniques. Experimental results show significant improvements by the two proposed techniques on the coding performance of the image compression algorithms.