Image segmentation technology based on partial differential equation and cluster analysis

Date

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Image segmentation is to extract objects that are interesting to users that is the most fundamental and important part in image processing. After decades of development and change, image segmentation methods based on various theories have been put forward. The active contour model based on the variational method and the level set method has become an important method of image segmentation. It demonstrates the superiority of partial differential equation image processing method. It mainly uses the idea of dynamic evolution, and has important significance in the study of image segmentation technique. For the study of the image segmentation based on partial differential equation, the numerical calculation good stability in the processing of discretizing the partial differential equations, and achieves high quality image restoration and accurate segmentation of images. In recent years, the active contour model has become a popular research method, and is widely used in edge detection, medical image segmentation and object tracking. This paper introduces the background, research status, the current development the purpose and the significance of image segmentation based on partial differential equations, the classical active contour model and the related mathematical theory. By studying the problems that exist in the image segmentation on the active contour model we established a new model of medical image segmentation based on information entropy. At the same time, we also studied the method of image segmentation based on clustering analysis (SFCM) and analyzed its own advantages and disadvantages. Furthermore, we established a fuzzy c-means clustering image segmentation model based on gray space. The main work and achievements of this dissertation are as follows: 1. We studied various methods and models of image theory of partial differential equations based on the segmentation, analyzed the theoretical basis and the mathematical principle of these methods, and compared their respective advantages and disadvantages. On this basis, we established two new models, and the numerical results show that the new models have prominent advantages. 2.When there is weak edge, strong noise, or uneven brightness in the image, the traditional active contour model can not achieve the correct segmentation of the target boundaries, especially for medical magnetic resonance images and ultrasonic images. To solve this kind of problems, we propose an active contour model CER based on CV model and RSF model combined with information entropy. By minimizing the energy functional and taking into account the internal and external energy of the target boundaries, the segmentation of the target boundary in the uneven intensity image is realized. The experimental results show that the algorithm can extract the object with weak edge and uneven brightness. At the same time, the RSF model’s robustness to noise is enhanced, and its sensitivity to initial contours is also improved. 3. Image segmentation method based on clustering analysis is one of the popular segmentation methods in recent years. The main feature of the traditional FCM clustering algorithm is that it is an unsupervised segmentation method with fast computation speed. However, it is sensitive to outliers. Therefore, we study an improved SFCM algorithm, whose biggest advantage is that it combines the spatial information of the image and reduces the sensitivity of the traditional FCM algorithm to noise. Based on this advantage, we propose to preprocess the image by SFCM, and to take the edge information of the image after clustering as the initial contour of the CER model. This method not only improves the CER model's robustness to noise, but also solves the initial contour selection problem. In addition, the rationality of the initial contour selection is improved and the computational efficiency is improved to a certain extent. A large number of experimental data show that this method can not only accurately segments the strong noise images, but also improves the initial contour selection problem and reduces the computation time, thus enhances the robustness of the CER model.

Description

Keywords

Citation

DOI