| In this thesis, the following problems are addressed: the design methods of orthogonal, biorthogonal wavelets, the design of the optimal filter banks for image coding, the methods of the boundary-processing before and after wavelet transform, efficient encoding of transform coefficients and et cetera. The key contributions of this dissertation are as follows.1. Based on the Daubechies' theory and design of compactly supported orthogonal wavelets, an fast implementation of the matched zeros-pair method for constructing compactly supported orthogonal wavelets has been produced. A general design algorithm has been proposed, which can generate any symmetric compactly supported biorthogonal wavelets with some regularity. Its software has been designed as well. With the algorithm one can easily obtain the spline wavelet filters and all of the important filters in application, of course many new filters can be obtained also.2. Two novel classes of filter banks have been proposed, which possess some optimal properties. They are NVMF systems and BNVF systems. Closed-form expressions for the impulse responses of the NVMF filters have been derived. A algorithm for constructing the impulse responses of the BNVF filters has been proposed. The two classes possess several remarkable properties. In particular, three FB's in NVMF systems and three FB's in BNVF systems have been shown to be competitive with the CDF-9/7 in DWT-based image compression. Furthermore, the multiplication-free DWT/IDWT using the NVMF filter banks are promising in the realization of real-time image and video code. The numbers of multiplications using the NVMF filter banks are about half of the CDF-9/7's in DWT/IDWT, its computational complexity is lower than using the CDF-9/7's.3. A new general M-Channel analysis/synthesis system with windowing and extending has been established. Two methods with nonexpansive property have been proposed, which is used for boundary-processing before and after multi-wavelet transform. They are best general symmetric extension method and r-factor pseudo periodic extension. It has been shown practically and theoretically that the two methods possess several remarkable properties, such as general application, flexibility and good compression potential.4. The relationship between zerotree coding and traditional run-length coding has been shown. In fact, similarly, the idea of zero-block coding is related to the one of run-length coding. Based on this, a new embedded zero-forest coding method MRZF has been proposed. Experimental and theoretical studies have shown that strong correlation exists not only among the significance of DWT coefficients of the same spatial orientation, but also among the zerotree locations across different directional subbands. Latter has not been exploited in any zero-tree/-block coding methods. Based on this, in MRZF encoding, 2-forest and 3-forest have been defined to represent the transform coefficients , which can effectively exploit all of the above correlations. The outputs are "0","1" bit stream, no entropy encoding has been included, MRZF encoding/decoding is very fast.5. Furthermore, the correlations also exist among the signs of adjacent coefficients, among the significance values of adjacent coefficients, adjacent forests and adjacent trees. The significance values of adjacent coefficients, adjacent forests and adjacent trees have been encoded jointly, and then the outputs have been arithmetic coded. So, the improved MRZF (IMMRZF) has been obtained. Both MRZF and IMMRZF have multirate, low complexity and progressive transmission properties.6. The whole process of image compression proposed in this thesis has been tested on popular standard test images at various rates. The experimental results have shown that the new method MRZF outperforms EZW+OWF[33] and zero-tree coder EZW[35], is close to SPIHT[36] and zero-block coder SPECK[108], in compression efficiency. The PSNR performs of IMMRZF proposed surpasses EZW[35] and SPIHT[36], and outperforms SPECK...
|
PDF Full Text Request