Texture Classification and Segmentation Based on Iterative Morphological Decomposition

Abstract

This paper presents a new method for texture classification and textured image segmentation based on grayscale mathematical morphology. It defines a recursive morphological decomposition algorithm by using a group of structuring elements with different sizes which decomposes a texture image into a series of component images according to texture primitive sizes and gray levels. Each component image contains only the texture primitives of a certain size, and the original texture image can be exactly reconstructed by the sum of all of its component images. Many texture features can be extracted from these component images for texture classification and textured image segmentation. This paper, then, proposes an adaptive textured image segmentation technique. The size of the window from which texture features are extracted is selected according to expectation of misclassification from the textures to be segmented. The window position for each pixel is determined by its neighborhood. This technique can improve segmentation results, especially along texture boundaries. The experimental results show that the method presented is fast in computation and efficient for classification and segmentation of both structural and random textures. Fairly good experimental results have been obtained, even though few features have been used.