Research And Application Of Active Contour Image Segmentation Method Based On PDE

Image segmentation has always been a classical problem in image processing.In many image segmentation algorithms,the method based on partial differential equation(PDE)is one of the most popular image segmentation algorithm and has attracted the attention of many scholars with its vitality and creativity.The basic idea of it is: first,build a mathematical model,called energy functional.Then the level set function is driven by the PDE which is obtained by minimizing this energy functional.The active contour moves at the same time and eventually stops at target boundary.Finally,the numerical solution of the PDE is the result of the segmentation.The famous models based on PDE include the GAC model,the CV model and the RSF model.Most of the subsequent models evolve from these models.In thesis,after analyzing the working principle of the classical model,the author of thesis has carried on segmentation experiments and put forward a new model and some improvements based on the latest research results,as follow:(1)Based on the construction principle of CV model,the author of thesis presents a feature description method based on the local complexity of the image and then constructs the natural image segmentation model with this description.The new description method transforms the complex natural image into a simple two-phase image,which leads to an expected segmentation result.It is not difficult to find that the new model can extract the texture target from the complex background accurately and quickly,while keeping the integrity of the target well.(2)The author of thesis analyzes the level set evolution model which is used to solve the problem of contour initialization and showed some improvements.In order to improve the robustness of the model to the noise and the iterative calculation speed,the bilateral filter is involved.Increasing the judgment item makes the improved model can be applied to the segmentation of the complex natural image.Experiments show that the improved model can improve the accuracy of segmentation,greatly reduce the number of convergence and fully extend the application field.