Research On Image Segmentation And Tracking Based On Level Set Theory

Image segmentation and target tracking are the prerequisites for image comprehension and recognition,which have been a hot topic in image processing and computer vision.Due to factors such as noise,low contrast,and uneven intensity,how to accurately segment the target object is still a challenging task.The difficulty of stable tracking the target for a long time is the target topology transformation,object occlusion,illumination variations,camera shaking and so on.The image processing method based on level set,has been widely concerned in recent years because of its advantages of free topology transformation and multi-information communion.Therefore,this thesis has carried out the research on target segmentation and tracking technology under the framework of level set theory.The main contents and achievements are shown as follows.Deeply research on image processing methods based on curve evolution theory and level set method.Introduce the basic concepts of curve evolution theory and the method of energy functional solution.Research on the basic principle of the level set method,the numerical solution of the level set evolution equation,and the promotion to the case of advective motion and curvature movement.Discuss and analyze the key technology of level set,such as the initialization of the level set function,the expansion of the velocity field and the selection of the iterative time step.Design a fast hybrid level set model.Fully analyze the characteristics of the classic level set model.In order to solve the problem of the sensitivity of the evolutionary curve to the initial position,and the segmentation problem of the uneven intensity images,new model combines the local image fitting term and the global image fitting term,to drive the contour evolution.The complicated reinitialization process is completely eliminated.The proposed method can accurately segment an image with a non-uniform property,and the final segmentation result is independent of the initial position of the initial curve.In the numerical implementation,the algebraic multiple grid is introduced to break the time step limitation,greatly reducing the time consumption of the evolution process,and making the algorithm quickly converge to the real location of the target.Propose a multi-feature fusion contour tracking method.The basic principle and classification of the contour tracking method are studied,and the performance of the classical active contour model is analyzed.The proposed method can obtain the discriminant model and realize the accurate extraction of the target by fusing LBP characteristics,HOG characteristics and HSI characteristics.According to the discriminant model,the confidence map of the target is obtained by the SVM classifier,and introduced into the level set frame as the external energy of the contour.The internal energy of the contour is added to guide the evolution of the contour to the edge position of the target.