Research Of Hierarichical Model Based On Student's T-Distribution For Image Segmentation

As a basic and siginicant tool of image processing,image segmentation is always highly valued by researchers.Over the years,there are thousands types of image segmentation algorithms,however,the image segmentation technology has not yet become a general theory.It is mainly caused by development of science and technology.Appearing more and more high-performance imaging hardwares,the acquired images become more complex and detailed,so,it brings a huge challenge for the entire image processing research field.With the development of many new theories and methods in various disciplines,the researchers began to try to apply the new theory and method to the image segmentation,resulting in many segmentation algorithms based on specific theories.In this paper,we mainly research the segmenting algorithm based on the clustering analysis,and propose a hierarchical algorithm model.And then,we use wo clustering algorithms——the finite mixture model and the fuzzy C-Mean,to verify the effectiveness of the hierarchical algorithm for image segmentation.The main work of this paper is as follows:1)Research and analysis of the basic knowledge of image segmentation technology,application background and current research status at home and abroad.In addition,we summed up and summarized the typical image segmentation algorithm,and analysis of these methods applicable scenes and shortcomings to provide direction for follow-up research.2)A hierarchical mixture model based on student's t-distribution(HSMM)is proposed.By analyzing the existing mixture model method,it is concluded that the traditional mixture model in sensitive to the local information of the image,is influenced by the image with contrast highly,and even produce the error segmentation.In this paper,a hierarchical model is used to fit the data and achieve a better effective for image segmentation.Secondly,we choose the student's t-distribution as the conditional probability of the mixed model,which is insensitive to the noise and outliers in the image.In addition,the mean template is introduced as the space constraint,which can make the partial information of the segmented image richer.3)A hierarchical fuzzy C-Mean(HFCM)model based on student's t-distribution is proposed.The Euclidean distance in the standard fuzzy C-means algorithm has no strong robustness to the noise,some existing methods have improved the standard FCM,and many more powerful and effective substitution functions are used to replace the distance function in FCM algorithm.In this paper,the hierarchical FCM algorithm using the strong robustness of the student's t-distribution to replace the European distance,we assume that the distance function is a sub-FCM estimate,each category of the distance function is composed of two or three subgroups of FCM,so can better approximate non-Euclidean distance function.In addition,the algorithm based on hierarchy can be extended to different distance functions,so our algorithm has better flexibility and versatility than standard FCM.