Texture Image Segmentation Based on Multiscale Analysis

In order to improve edge accuracy and regional consistency of texture image segmentation, this paper presents a multi-scale wavelet analysis as a tool, to improve the texture segmentation method by reducing the affect of segmentation resulting from redundant features. Three stages-feature extraction, optimization of the eigenvectors and clustering-are included in the method. In the stage of the filter effective features, by directly remove dimensions of redundant features which are more grouped, the features of dimensions which filtered out are given to determine the appropriate weights to optimize the feature vector. Compared with traditional methods, when texture images are segmented, the method can significantly improve the accuracy of edge, region homogeneity and segmentation error rate.