Wavelet Transform In Image Compression Encoding

This dissertation is about the application of wavelet theory in image compression. Wavelet is the grand new stage of harmonic analysis. Recently wavelet has become popular in image and video applications.Based on the reviews of compression principles, coding methods and image information theory, we indicate that the essence of information is not only statistical, but also structural. We name structural information as semantic information. The property of semantic information is conjectured, and Shannon's communication model is extended. Semantic information coding should utilize the property of semantic information channel.Wavelet theory is discussed in the clue of discretization and information representation. Wavelet provides a compact multi-resolution representation and reconstruction of image, which makes it possible to exploit the property of Human Visual System for image coding. In this paper a coding method is presented based on wavelet, self-organized mapping neural network and vector quantization. Firstly, the image is decomposed in multi-resolution by wavelet transform, and then we encode the low frequency scale image using DPCM, simultaneously the high frequency wavelet coefficients is trained by self-organized mapping network to form a code book, and then encode the quantization results. In order to improve the generality of the code book, we embody adaptive resonance architecture to balance the stability-plasticity trade-off of the quantization system. With the consideration of the property of adaptive resonance theory and wavelet vector quantization, a new way, which simplify the adaptive resonance architecture network and algorithm, is put forward. The method provides a mechanism that gradually matches the signal's property and improves the generality of the code book. Experiment results prove the efficiency of the method.The multi-scale edge detection and reconstruction of image using wavelet validate the structural essence of image information. We achieve the multi-scale edge detection results of image adopting spline wavelet and Mallat algorithm, then introduce a semantic information coding method applying the structural multi-scale property of image.