Research On Image Segmentation Method Based On Non-locally Anisotropic Structure Tensor

As a basic task in the field of image processing,image segmentation is also one of the three major classification tasks in the field of computer vision,which focuses on the classification of each pixel in the image.It has been widely used in remote sensing,medicine and other fields.However,for SAR image segmentation in remote sensing field,the result is often affected by speckle noise,which makes the quality of SAR image segmentation decline.Secondly,in the process of medical image segmentation,it is often affected by a variety of artifacts,such as noise,uneven signal strength and so on,and ordinary images are often affected by gaussian noise in the process of shooting and transmission,which leads to the degradation of image segmentation performance.To solve the above problems,this paper focuses on the application of nonlocal anisotropic structure tensor in various segmentation methods.The main work of this paper is as follows:1)A hierarchy student's t-mixture model based on nonlocal anisotropic structure tensor is proposed.An adaptive weighted template based on nonlocal anisotropic structure tensor is introduced into the student's t-mixture model to obtain more prior knowledge and enhance the robustness of SAR image segmentation.Secondly,a hierarchical student's t-mixture model is proposed,which divides the global SAR image segmentation problem into several sub problems.On the one hand,the complexity of the problem is reduced.On the other hand,the hierarchical idea can make the model better fit the data.Finally,by comparing with some existing image segmentation methods in SAR image data set,it shows that the algorithm has better performance and stronger robustness in SAR image segmentation.2)A fuzzy c-means clustering algorithm based on nonlocal anisotropic structure tensor is proposed.As for the fuzzy c-means clustering algorithm,a smoothing template of nonlocal anisotropic structure tensor with the ability of image content analysis is introduced to ensure that the image details are preserved in the process of segmentation and better deal with all kinds of noise,so as to overcome the problem of poor segmentation effect of fuzzy c-means clustering for noisy images.Secondly,using the initial clustering point selection algorithm based on image gray histogram fitting,the initial points and clustering number are obtained in advance through the student's t-mixture model,which can solve the problems of manual definition of clustering number,too many iterative steps and slow convergence of objective function caused by improper selection of initial points,so as to speed up the segmentation speed.Finally,the proposed method is compared with some existing image segmentation methods in medical image and common image datasets to verify that the proposed method is more robust in the segmentation of noisy images.