基于不完全树结构小波变换及模糊聚类神经网络的纹理分割

TEXTURE SEGMENTATION METHOD BASED ON INCOMPLETE TREE STRUCTURED WAVELET TRANSFORM AND FUZZY CLUSTERING NETWORK

提出了一种新的不完全树结构小波变换用于纹理特征提取 ,给出了一种与人类视觉过程相一致的多分辨率多通道纹理分析方法 ,它由 :1)特征提取 :使用不完全树结构小波变换抽取纹理特征 ;2 )基于模糊神经网络的特征粗分类 :1基于样本分布密度的模糊 Kohonen聚类网络权值初始化 ,2使用缩减的特征向量对网络进行训练 ,得到粗分割结果 ;3)细化粗分割结果等几部分构成 .

A new type of incomplete tree structured wavelet transform used for texture feature extraction is proposed in this paper, a new texture segmentation method consistent with human visual process and based on incomplete tree structured wavelet transform and fuzzy clustering network(FKCN) is proposed also in this paper. It consists of three main steps:1)feature extraction using incomplete tree structured wavelet transform;2)feature coarse classification using fuzzy Kohonen clustering network: before the training process, the weights vector of the network is initialized according to the density function of input patterns, then network is trained using the reduced set of feature vectors and get the coarse segmentation result;3) refine the coarse segmentation result. After refinement of the coarse segmentation result, the final segmentation result is obtained. Texture segmentation experiments show the effectiveness of this method.