Change The Level Set Method Based On The Overall Curvature Surface Recovery

The 3D reconstruction based on CT images, Point-Cloud and Multi-view images have become the main methods in computer vision field. However, the surface given by every method has noise and disrepair phenomenon inevitably which can influence the study followed. So, it is important to do the research of denoising and inpainting. Surface denoising and surface inpainting are studied in this dissertation. The main contents are as follows: Firstly, the ROF model based on the TV flow is introduced, meanwhile the edge-preserving characteristic of TV flow is studied. The staircase phenomenon of ROF model is analyzed, and the characteristics of Bing Song adaptive model and its property of edge diffusion with respect to different region of adaptive parameter are also studied. Secondly, the level set method is introduced, and the C-V model is analyzed. Then the variational level set method is extended from 2D into 3D field which supplies the research object for the following study. Thirdly, for surface denoising, a model based on total curvature and variational level set methods is proposed which is the analogue of total variation based on ROF model for image denoising. The model based on the implicit variational level set methods can handle topological changes automatically and can not only denoise effectively but also preserve edges and key details of the surface. An adaptive denoising model based on total curvature is proposed later due to the adaptive parameter which can realize the different edge diffusion degree of the surface. Finally, a surface inpainting model is proposed by extending the denoising model into inpainting filed. Experimental results show that the denoising model removes noise efficiently and preserves salient features of the surface and the inpainting model which can inpaint the small damaged region can handle simple surface inpainting problems well.