Image segmentation using fuzzy homogeneity criterion

Abstract

The approach proposed here is using fuzzy homogeneity vectors and the fuzzy cooccurrence matrix to segment images. Homogeneity vectors could be used to represent the homogeneous feature between a pixel and its neighbors, and the degree of homogeneity could be determined by using a fuzzy membership function. Combining the homogeneity vectors and the fuzzy membership function, we could extract the feature of an image which determines the fuzzy region width. With the fuzzy region width, we could use the fuzzy cooccurrence matrix to measure the fuzzy geometry properties of an image, which is the fuzzy entropy values that can be employed for determinine the thresholds to segment images. The advantages of this approach are as follows. First, homogeneity vectors take account of the spatial gray-tone dependence; thus, using homogeneity vectors has a better noise tolerance. Second, from the extracted feature, the fuzzy region width could be determined automatically. Third, because the fuzzy region width is decided by the feature of the image, it could be adjusted according to the nature of the image. Fourth, the dimension of the fuzzy cooccurrence matrix representing the image is significantly decreased, but the properties of images are faithfully preserved. A large number of experiments have been carried out on different kinds of images, and good results have been achieved by the proposed method. This method will have wide application in image processing.