The Improvement Of Image Segmentation Methods Based On Active Contour Models And Their Applications

As one of the most important steps of image preprocessing, image segmentation drew more and more people’s attention. In these years, image segmentation method based on PDEs, especially based on active contour models, is a vital technique in image segmentation and has a wide application due to its effectiveness and automation. In the area of image segmentation, the methods based on active contour models is characterized to Chan-Vese model provided by Chan and Vese and the geometric active contour model based on variation put forward by Chunming Li, which used variational level set methods. respectively. Many improved PDEs models were based on these two methods. In this paper, two improved methods without initialization are proposed based on the models above:1. A new level set model based on PDE is proposed to enhance the segmentation ability for images with full details or with complex boundaries. The new model uses a outside force with the form of Laplacian to enhance the importance of boundary in the evaluation. On the other hand, the new model uses the internal energy to restrict the level set function which can avoid re-initialization in numerical computing and improve the computing efficiency. Experimental results show that, the segmentation effect and the running time of the new model was much better than the two models above-mentioned when processing noisy images and images with complex edges.2. The C-V model only uses the grey intensity information to build segmentation model which cannot recognize each object in multi-target images. In order to address this problem, the energy items based on image gradient are used to improve the Chan-Vese model; and the Euclidean norm of the gradient of the level set function is used to replace the regularized Dirac function in the Chan-Vese model for keeping segmentation stability and eliminating the restraining of Dirac function. The experimental results show that the segmentation results by the proposed method in this paper are better than the two models above-mentioned when processing images with "hole" and "thick" edges, multi-target images.The segmentation results of pavement surface distress image show that, the porposed models are practical, and can be used to recognition systems on the pavement surface distress image.