Texture features extraction in the fuzzy Kohonen neural network for image clustering

Nahla Ibraheem Jabbar

Physical Sciences Research International
Published: February 26 2014
Volume 2, Issue 1
Pages 24-27

Abstract

In this paper, a new approach of fuzzy Kohonen neural network (FKNN) for image clustering using texture features of biopsy images was studied. The statistical textures have been extracted from co-occurrence matrix. This method allows the computing of different types of features: Energy, Homogeneity, Entropy and Contrast. Statistical texture can be defined as the spatial distribution of intensity variation in an image according to some underlying probabilistic model. These features are fed as input into the fuzzy Kohonen neural network (FKNN) to specify the number of clusters in the image. This technique is an example for estimation the number of clusters from searching a set of texture features to finalized image segmentation by fuzzy Kohonen neural network. Our proposed method is used to overcome the problem of large data set of pixels values as an input to fuzzy Kohonen neural algorithm to the number of texture features. These texture features are represented by a pre-processing step in the estimation number of clusters by FKNN. Our experiments show the ability of the proposed algorithm in the image segmentation with a quality number of texture features. It gives a high data clustering efficiency. This algorithm works effectively after noise has been removed from the image.

Keywords: Texture features, clustering, fuzzy Kohonen neural clustering, fuzzy c-mean clustering, segmentation.

Full Text PDF