FUZZY GEOMETRIC FEATURE-BASED TEXTURE CLASSIFICATION

It is necessary to compute various spatial and gray-level properties f or the classification of textured regions of an image. However, the re gions and their properties are not always crisply defined. It is more appropriate to regard them as fuzzy subsets of the image. In this pape r we have proposed the use of fuzzy geometric properties for texture c lassification. At first, a set of 2-D local membership-value extrema h ave been detected on the image. Using them as 'seed' regions, they are grown till the grown regions do not touch any other seed regions. The resulting regions are called the regions of influence. Fuzzy geometri c properties like fuzzy area, perimeter, compactness, height and width are determined on these regions and they constitute the feature space for texture classification. Several natural textures are digitized an d used to test the efficiency of the approach. It is seen that about 9 0% classification accuracy is obtained in the pattern space of 8 textu res.