Study On Image Segmentation Based On Textural Feature

Nowadays Texture image segmentation is a hot research topic, because it isimportant for image processing, pattern recognition and computer vision. Textureimage segmentation means that an image which is composed by different textures isdivided into several regions which have the same or similar texture. Texture imagesegmentation contains two parts: feature extraction and region consistencysegmentation algorithm. The result of Texture image segmentation is a set whichcontains labels that are used to label every pixel.This dissertation is devoted to where the number of texture patterns is knownbut the information about their properties is not. After comprehensively reviewingthe basic principles and existed methods, the author chooses the feature-basedapproaches to solve this problem. Generally, feature-based texture segmentationalgorithms can be viewed as consisting of two successive processes: featureextraction and feature partition. In this dissertation, the author investigates thosetwo processes, respectively, and achieves highly effective, increasingly innovativeand cutting-edge approaches of texture segmentation, which can be summarized asfollows:1. Described in detail based on GLCM, Gabor filter texture feature extractionalgorithm, and describes some of the improved algorithm.2. A new group of texture features are presented in this paper: based onGeneralized Local Walsh Transform(GLWT)and Local Binary Pattern(LBP). Thedefinition of GLWT is given. Then Calculate and analysis the GLWT coefficients.And the2ndorder moment are selected as texture features. Combined with LBP andthe gray space Characteristics, then all the pixels are clustered by fuzzy C-meansalgorithm. The experimental results reveal that the texture features presented byCombined with GLWT and LBP have the best discriminating performance, andsimple calculation.