A Research Of Texture Image Segmentation Based On Extracting Features

Texture image segmentation is one of research emphasis in the field of image processing;it has a very important position in many fields such as computer vision, pattern recognition. Itsegment texture image by the consistency of same texture and the differences betweendifferent textures. Texture can be described by features; using features can show theconsistency of same texture and the differences between different textures. Therefore, the keyto texture image segmentation is two problems: how to extract texture features and how toanalyze texture features. This paper is researching how to solve these problems and doingtexture segmentation. The specific research work are as follows:First, this paper analyses the Gray Level Co-occurrence Matrix(GLCM) which is one ofclassic texture feature extraction algorithm, start from the basic principle, analyses the role ofeach parameter which can build GLCM, do experiments to sure the result of extractingfeatures by different parameters, find the best value range of parameters and the deficiency inGLCM;Second, it analyses the Local Binary Pattern(LBP) and its different forms, fuse with LBPoperator and GLCM to extracting texture features, proving the validity of this method byclassifying texture image, to determine the values of parameters, and overcome someproblems which is caused by the value of parameter in GLCM;Third, this paper introduces the principle of Fuzzy C-means clustering algorithm (FCM),using this algorithm to analyse the texture features;Fourth, this paper presents a texture segmentation method. The method fused with LocalBinary Pattern (LBP), Gray-level Co-occurrence Matrix (GLCM) and permutation entropy. Itfuse with uniform pattern LBP and GLCM to extract texture features, angular second moment,contrast, energy and inverse difference moment as part of features, then, build a feature vectorby these four features and the permutation entropies calculated in many different directions.The method use FCM and spilt-and-merge method to anlyse features, the center of ahomogenous texture is analyzed by using features in coarse resolution, with a proper strategy,transmit the result to finer resolution and find the border of different textures, then anlyse thefeature of border in finer resolution, unites with the result in coarse resolution, to get the result of texture segmentation. Experiments on texture image demonstrate that the proposed methodleads to a successful segmentation, and give the evaluation of the accuracy.