基于RBF小波网络的图像预测编码-计算机应用技术专业论文.docxVIP

ABSTRACT An image’s main energy and features concentrate on wavelet coefficients at low frequency subband after wavelet transformed．It’s difficulty to compress the low frequency wavelet coefficients in image coding algorithm based on wavelet transform But it does not indicate that these coefficients are not to be compressed．The experimental results prove that the neighboring rows of the low frequency coefficients are highly correlated and the common predictive coding algorithms can not make full use of the relativity．And because the RBF neural network is able to approach an arbitrary nonlinear function，So an image predictive coding method based on RBF neural network is presented in this thesis．This coding method can utilize the relativity well and reduce the redundant data． The lifting scheme biorthonormal wavelet is constructed and used to transform the experimental image in this thesis because of the merits of lifting scheme wavelet and biorthonormal wavelet．It is a crux to construct an appropriate RBF neural network in the predictive coding algorithm．There are two ways to construct an RBF neural network in this thesis．One means is to construct an initial RBF neural network by myself，then optimize its initial Structure by improved learning algorithm based on self—structure．Another means is to construct an RBF neural network by using neural function in MATLAB tools box．After comparing the running performance of the networks constructed by two means，the interface means is better,so it is used to construct RBF neural network in image coding experiments． The image coding experimental results indicate that the constructed RBF neural network has rough predicting function and adapting function．The image coding method presented by this thesis can provide higher compression ratios than DPCM coding，EZW coding and Huffman coding in image coding experiments．So the predictive coding method based on RBF wavelet network is a very useful and effective image coding method． Key wo