Research On Gray Image With Markov Random Field

Along with electronic equipment widely used, the popularity of social networks, the development of telemedicine, more and more digital photo have been produced every day. Image segmentation is the first step of digital photo process. If image segmentation failed, we cannot do the next step. So far, we have many image segmentation techniques, such as threshold method, edge detection, regional method, based on mathematics image segmentation, based on pattern recognition image segmentation. Today we can find that based on mathematics image segmentation is very popular used. By mathematical statistics we can find and extract information. Based on mathematical statistics have been widely used in image segmentation, such as K-Means, C-Means, MRF (Markov Random Field). Markov Random Field is good at describing the spatial context of photo, and it is widely used in image segmentation.In this dissertation, we will use the Markov Random Field to improve the efficiency of fuzzy C-Means Algorithm, using the improved fuzzy C-Means algorithm to segment the known quantity. We can find out that the improved fuzzy C-Means have good result. In the real life, there are many photo we cannot give the number of image segmentation, and we can’t use the improved C-Means to segment. In this dissertation we introduce the HCF algorithm to solve this problem. But in the HCF algorithm, there are over global question. So we put forward the improved HCF algorithm, it solve the over question. We can find out that improved HCF algorithm have good segmentation efficient as well as concurrency. This dissertation works include:First, introducing status quo of image segmentation. Simply introduce the principal of threshold method, edge detection, regional method, based on mathematics image segmentation, based on pattern recognition image segmentation.Second, introducing the theory of Markov Random Field. Simply introduce:1. the concept of neighborhood and clique system in Markov Random Field.2. The common model of Markov Random model.3. Compute the parameters in Markov Random Field.4. The optimal image segmentation in Markov Random Field.Third, adding Markov Random Field to C-Means. C-Means have been widely used, but the membership and distance in C-Means have great impact on effect. In this dissertation we add the spatial feature of Markov Random Field to C-Means. Conditional probability of Markov Random Field replace the membership and distance in C-Means. We do experiment in noised image, texture image and CT image.Fourth, improving HCF algorithm. In HCF algorithm we needn’t to give the number of image segmentation. HCF algorithm label every point as edge or non-edge. If one region is closed, we say it is one class, and we find out the class number of the image. In HCF algorithm, the global optimal may be over. In this dissertation, we use the local optimal to replace the global optimal. The experiment result show that our method precede in time and image segmentation.Fifth, making a summary of this article and proposing future work.