The Research Of Texture Image Segmentation Algorithm Based On Fuzzy C-means And The Coupled Hidden Markov Random Field Models

As is well known, texture exists in much natural scene. Texture segmentation is a significant branch of digital image processing and plays an important role in computer vision. However, due to the diversity of images, the complexity of natural textures and the lack of understanding of human vision system.The segmentation of textured image is still a major problem in image processing. During the past four decades, hundreds of segmentation algorithms have been proposed in the 1iterature.But, those algorithms usually suffer from less accuracy and narrow image specific orientation. Therefore, texture segmentation is, up to now, still an open topic with great challenge in image processing field.This dissertation is devoted to the segmentation of textured gray level images, after comprehensively reviewing the basic principles and existed methods, the author chooses the feature-based approaches to solve this problem. Generally, feature-based segmentation algorithms can be viewed as consisting of two successive processes: feature extraction and feature partition. In this paper, the author investigates those two processes, respectively, mainly done the follow works:(1) Because the clustering algorithm does not require the provision of training samples, is a non-supervised statistical methods. Owing to the fuzzy C-means clustering algorithm dose not consider the images of the spatial correlation. Hidden Markov random field(HMRF)models, taking into account the mutual influences of neighboring sites, have been widely used for image segmentation. This dissertation ,we combine the benefits of these two approaches, proposed a new image segmentation methods based on the FCM-type clustering and the HMRF, which by FCM clustering iterative we can easy obtain HMRF model parameters.(2) Texture is a regional characteristics, texture feature extraction must in an area. According to the principle, A coupled texture feature extraction is proposed in this paper. The method has two mutually dependent components: one model is the observed image to estimate feature, we adopted the Gaussian Markov random field, the other model is the labeling to achieve segmentation. When calculating the feature of each pixel, the homogeneity of the sub-image is ensured by using only the pixels currently labeled as the same pattern, with the acquired features, the labeling can gain more accurate label.The proposed approach is compared with other classic algorithms in segmentation of synthetic, brodatz texture mosaics and medical images. The satisfying experimental results demonstrate that the proposed approach can differentiate textured images more accurately.