Segmentation of textured images using gibbs random fields

Abstract

In this thesis digital image is introduced as an amount of data, which is produced when a 2-D light intensity function is sampled and quantized to create a digital image. Its principle objective is to define the segmentation process, in which we can partition an image into meaningful regions that correspond to part of, or the whole of objects within a scene. This is done by systematically dividing the whole image up into its constituent areas or regions. If the regions do not correspond directly to a physical object, or object surface, then they should correspond to some area of uniformity. Major approaches to segmentation have been introduced. However, since texture plays an important role in image analysis and understanding. As a front end in a typical classification system, texture feature extraction is of key significance to the overall system performance .There have been many papers, proposing different approaches to this problem. This thesis also outlines the basic methods of texture analysis and comments on different issues. It also point to the connections of related methods. Texture segmentation is also introduced. It is an interesting but difficult problem in image processing. The main difficulty of traditional texture segmentation is the lack of adequate tools to characterize different scales of texture effectively. Recent developments and researches help to overcome this difficulty. We present many approaches such as the Quad Tree Segmentation, Gabor wavelet scale for Unsupervised Texture Segmentation, Unsupervised Texture Segmentation Using Multiresolution Analysis For feature Extraction, and Markov Random Field Model and Segmentation. The textural features are extracted from each decomposed image. The procedures results in a segmented image whose regions are distinct from one another with respect to texture characteristic content. Finally, a new approach was presented to the use of Gibbs distribution (GD) for segmentation of textured images. Specifically, random field models were presented for textured image data based upon a hierarchy of GD. Then the dynamic programming based segmentation algorithms for textured images were presented, considering a statistical maximum a posteriori (MAP) criterion. Since the model parameters are needed for the segmentation algorithms, a new parameter estimation technique is developed for estimating the parameters in a GD.