Mesh Denoising Via Guided Normal Filtering

Mesh denoising is one of fundamental problems in computer graphics community. With the widespread of 3D scanning devices and the development of 3D modeling tech-niques, more and more 3D mesh models are available. Due to the imperfections of raw data obtained by 3D scanning devices and the errors that surface reconstruction algo-rithms produce, some models are inevitably contains noise and outliers. Preprocessing including denosing of such models before they can be used is particularly important. Mesh denoising is intended to obtain high-quality 3D models from corrupted mesh data, i.e. removing noise and outlier while preserving geometric features such as sharp edges and corners. Recently, researchers have made significant progress in mesh denoising, such problem still remains challenging in several aspects:first, it can be difficult to distinguish geometric features from noise since both of them are of high frequency; second, many mesh denoising algorithms come from their counterparts in image denois-ing. Due to the irregularity and non-uniform sampling of the surface, mesh denoising is more complicated compared to image denoising. Third, various and different scale noise may co-exist in a mesh model. All of above imply that special considerations should be taken in developing the mesh denoising algorithms.In this dissertation, we propose a novel mesh guided normal filtering algorithm based on the joint bilateral filter to overcome these challenges for mesh denoising. The new algorithm is designed as a two-stage process:first, we apply joint bilateral filtering to the face normals; afterwards, the vertex positions are updated according to the filtered face normals. At the heart of the new algorithm is to construct a proper normal field as the guidance to adapt joint bilateral filtering from two-dimensions image to mesh in three dimensions. In particular, we compute the guidance normal on a face using a neighboring patch with the most consistent normal orientations, which provides a reliable estimation of the true normal even with a high-level of noise. The effectiveness of our approach is validated by extensive experimental results.