Research On Image Segmentation Based On Local Region-based Active Contour Model

Image segmentation is the foundation of image technology. The first step of image processing, analysis and understanding is to extract target region from the original image through image segmentation, then the further processing and analysis can be done for the image recognition or understanding. Active contour models have been widely applied in image segmentation without prior information for promising results.However, most of the active contour models fail to segment images successfully due to intensity inhomogeneity. In order to solve this problem, this paper focus on the image segmentation methods based on active contour model, the basic principle of active contour model and a detailed analysis of the advantages and disadvantages of various models is introduced. The local region active contour model is introduced in de-tail, and a hybrid region-based active contour model which fully utilizes local intensity information is proposed, the specific work is as follows:(1) A novel localized active contour model which incorporates a local Gaussian distribution fitting energy is proposed in this paper. In image segmentation, the local-izing region-based active contour model is effective but he accuracy of the model is not satisfactory, while the local Gaussian distribution fitting model has high stability in noise conditions and is not sensitive to the initialization of active contour but it is too sensitive to small intensity differences. Combining the advantages and disadvantages of the two model, we propose a hybrid local region-based active contour model, which can ensure the correctness of the segmentation results and improve the accuracy. The experimental results show the superiority of the proposed model.(2) Two kinds of optimization skills are designed and implemented in level set method. In view of the problem that the active contour needs to be re-initialized, reac-tion diffusion is used in the level set method of the active contour model for avoiding the re-initialization. This method not only accelerates the evolution of active contour model, but also ensures the robustness of the proposed model. In the iterative process of the level set, the narrow band method is used to further reduce the computation cost and improve efficiency.(3) A multi-resolution framework is applied to the hybrid active contour model. In order to reduce computational complexity, a collection of images with different resolu-tion level is constructed, and each picture is segmented with the proposed hybrid model. In the case of low resolution, the computational complexity is reduced, the motion range near the boundary is increased, so the number of iterations is reduced for speeding up the evolution of the active contour model.(4) A new initial boundary conditions is considered to the proposed model. Due to the complexity of the input images, the initial boundary is no longer a curve but two initial curve, the inner boundary and the outer boundary. These two curves constitute the action domain to constrain the position of the evolution curve better, and can obtain better segmentation effect.