Study Of Texture Segmentation Methods Based On Fuzzy Clustering And Neural Networks

Image segmentation is the basic and the most important step in object recognition~ image interpretation and computer vision. The segmentation result has strong influence on the following recognition and interpretation, therefore, image segmentation has been paid special attentions by many researchers. But up to now, it has not been resolved well. The research of human vision mechanism and the rapid development of computer techniques will provide new ways to solve this problem. The research of human vision system show that the spatial/frequency multi-resolution analysis based image segmentation method is coincide with human vision characters, it reveals the multi-resolution? multi-channel characters of human vision perception process, therefore, the spatial/frequency multi-resolution analysis based image segmentation method has drawn extensive attentions of researchers. Fuzzy set theory and artificial neural network image processing is a new research field with rapid development in recent years, and the fuzzy set theory and artificial neural network based image segmentation method become an important way of image segmentation. Although many works have been done in this research area, there are still have a long way to go to get a satisfactory result. This paper maid.~studies the effective texture segmentation method which consistent with human vision perception characters. The texture segmentation methods in this paper are combination of human vision perception characters, fuzzy set theory and artificial neural network. The research work of this paper can be classified in the following four respec~ First, the multi-resolution multi-channel texture segmentation model and texture feature extraction method have been studied. An effective texture segmentation model has been proposed, and the texture feature extraction methods by using 2D optimal polar separable orientational filter and wavelet transform have been given. The problem of local statistical feature values refinement has been discussed, and a refinement method has been given which can decrease the boundary effect in calculating the feature values. Second, the fuzzy clustering algorithms have been studied and the texture segmentation based on them have been given. The fuzzy C means and semi-fuzzy C means algorithm have been discussed. The similarity measure in fuzzy C means algorithm has been extended, and a semi-fuzzy C means algorithm based on revised Euclidean distance has been proposed. A 3 Abstract partially supervised fuzzy C means algorithm is proposed dealing with the drawback of partially supervised fuzzy C means algorithm need to determine the weight value beforehand and not reflect the samples?geometry characters, and the algorithm also has been used in texture segmentation. The third, the fuzzy neural network has been studied, and the neural network based texture segmentation method has been given. The fuzzy Kohonen clustering work (FKCN)has been discussed. By introducing the concept of A cut set and æ…ºuzzy decrease?operator into the learning rate function, two improved FKCN (IFKCN) are gotten, and the effective texture segmentation method based on IFKCN is given. An error measure weighted fuzzy vector quantization algorithm is given and used in texture segmentation. The multi...