Research On Embedded Image Coding Based On Wavelet Transform

Wavelet theory has been a topic of research in application math and engineering science. A typical application of wavelet is in image compression, and there are a few significative wavelet image codec. Embedded image coding is a new coding method and can provide the possibility to progressively reconstruct better and better reproductions of the image as more and more of the coding image bit-stream is received. Embedded image coding is especially fit for the changeful and fallible environment, such as internet/wireless transmission, image browse etc. The output bit stream of the embedded code can be truncated and reconstruct image for practical applications to achieve varying degrees of scalability and different target bit rates or distortion. This dissertation develops the different correlation of wavelet coefficients and suggests two significative embedded image coding by means of some characters of wavelet coefficients.The main work of this dissertation is summarized as follows:Continuous and discrete wavelet transform is researched according to analysis of the origin of wavelet transform and the shortcoming of Short Time Fourier Transform. Mallat algorithm is deduced from the viewpoint of Multi-Resolution Analysis. Filter banks is used to construct orthogonal and biorthogonal wavelet basis.Wavelet basis is selected according to the requirements of image compression. Finite support signal is transformed by wavelet basis in order to reconstruct image as possible as alike. Experiments show the distribution character and the degree of correlation of sign, within-subband and cross-subband of wavelet coefficients and provide transcendental knowledge for later coding.Context-based sign coding is used to remove the redundancy of sign of wavelet coefficient for correlation of them. Predictive coding based on the grads of neighbor coefficient is proposed to encode LL subband because LL subband occupied a majority of energy of transform coefficient. Adaptive run length coding explores the residual redundancy of the high frequency coefficients in the transform domain. Hereby, an embedded image coding algorithm based on Binary Description Run Length (BDRL) is brought forward.Morphology Classification based Gray Structuring-elements Dilation (MCGSD) is another embedded image coding to obtain farther excellent compression performance. It's reviewed for the coefficient classification of existing wavelet image encoding technologies. Wavelet coefficient is clustered with gray structuring elements dilation and outputted in the form of fractional bit plane according to the different significance. Zero tree structure is used to employ cross-subband dependency topromote the coding effort. The dilation blindness is avoided by seed coefficients with the help of knowledge encoding on-line.The main innovations of this dissertation are summarized as follows: Bring forward respectively coding the magnitude and sign of the predictive error of LL subband, which can combined some existing wavelet image codec to improve result. Proposing to adaptively encode run length with 3 symbol's arithmetic coding in term of it's binary form, which can be generalize any codec using run length. Hereby, BDRL is realized to form embedded bitstream. Introducing gray structuring elements dilation to the classification of wavelet coefficients, producing different significance for different clustering and benefiting for fractional bitplane coding. Suggesting decompose encoded LL subband to get seed coefficients of dilation firstly. At the beginning of each bit plane coding the seeds is acquired ,by the within-subband and cross-subband dilation of over significant pixels and significant pixels. Thereby the dilation blindness is avoided. Using zero tree to employ cross-subband dependency and. improving traditional cross-subband method of morphology compression algorithm. Implement an embedded image coder which is MCGSD having the functions of progressive transmission, arbitrary truncation etc.