Texture segmentation using competitive learning algorithm with pyramid approach

Abstract

In this paper, we propose a new segmentation method of an image composed of some kinds of textures by Walsh spectrums and competitive learning with pyramid approach. After a texture image is divided into nonoverlapping small windows with the same square size, the texture feature vectors in those windows are extracted by Walsh spectrums. In this paper, we propose a simple competitive learning (SCL) with pyramid structure that has one pixel on a higher level can be cluster number of n*n square pixels on a lower level which has a higher resolution. The clustering of feature vectors is performed by simple competitive learning algorithm and then the candidate clustering numbers are obtained. These cluster numbers make a new input vectors. These vectors are presented to the neural network of SCL as input patterns again. Like this, SCL is applied recursively like this until closure measure is satisfied. The simulation results show that misclassification rate is decreased in our proposed method.