Image segmentation is one of the main issues in the computer vision and image processing. The task of image segmentation is to divide a given image into multiple segments according to attributes of some distinctive visual features including brightness, color and texture. One of the challenging issues for image segmentation is to segment images with using textural features. There are several strategies for segmenting images and a lot of different segmentation methods can be found in literature. Normalized cut is a famous graph-based image segmentation scheme which focuses on pixel’s gray-level without any attention to other attributes such as texture features. The aim of this thesis is to improve the performance of normalized cut for segmenting textural images. The first step of improving the quoted method in the exposure of textural images is the use of a proper texture descriptor. We analyzed effects of different texture descriptors and finally came up with the suitable one. Moreover, to decrease complexity and increase efficiency of the proposed method, a homogeneous group of pixels are used in the computation as a super-pixel. In the second step, by applying the segmentation operation in two phases and delay of the decision in case that cannot make proper decision, the segmentation results will be improved. To compare the performance of the proposed method and some well-known similar works, the proposed method has been applied to the Prague texture segmentation benchmark. The experimental results indicate that the proposed method has better performance in comparison with the considered works. Keywords: Texture segmentation, Graph-based image segmentation, Super-pixel
