Research On Mesh Processing With Feature Preserved

With the development of3D scanner and modeling technology,3D geometry mod-els are becoming more and more popular which are widely used as an emerging type of digital media in computer animation, computer game, industrial and mechanical de-sign, computer simulation,3D visualization, medical diagnosis and so on. In the past decades, we have witnessed significant advances in the modeling, rendering and edit-ing of3D meshes. In the mesh processing, acquirement of salient features is a basic task. Salient features are important regions which include distinctive information and can guide the following mesh processing. The previous methods focus on using purely geometric measures such as normal and curvature to find the features. But a purely geometric measure may fail to find correct regions in many cases. For example, region-s with high curvature are usually regarded as important ones by traditional methods. However, densely repeated details, even if high in curvature, are suppressed by the hu-man visual system. A flat region in the middle of repeated high-curvature bumps will be perceived to be important.Mesh saliency is a new perception-inspired metric for regional importance. The image saliency has been studied extensively, and used widely in object recognition, image retrieval, image segmentation and editing. However, there are few studies on mesh saliency. The computation of mesh saliency is a challenging job, because the mesh models have more complex information than images such as vertex, topology, normal, curvature, texture and so on. In this thesis, we have performed an indepth study on mesh saliency, and discussed its application in mesh simplification and mesh resizing. The main contributions are as follows:(1) In the previous methods, only local contrast is used to compute the mesh salien-cy. We propose a new method for saliency computation, which considers both local contrast and global rarity. According to the center-surround mechanism of human vi-sual system, the method performs a Gauss convolution on mean curvature of mesh, and uses multi-scale DoG (Difference of Gaussian) operator to compute the local contrast. To improve the computational efficiency, the method applies a simple clustering method on the mesh vertices based on Gaussian filtered curvature, and compute the global rarity of every vertex cluster to other vertices. Local contrast and global rarity are combined to get the final mesh saliency map.(2) Mesh simplification is to reduce the geometry elements of mesh, while trying to preserve its important features. We apply the saliency map to the mesh simplification, and propose a new simplification method based on edge collapse. When computing the collapsing cost, the method considers both geometric measure and visual saliency value. The saliency value works as a weight, and makes the salient edges have bigger collapsing cost. The method preserves the salient features well in the processing of simplification, and gives more visually appealing results.(3) Non-homogeneously mesh scaling will resize the model in different directions, and leave the vulnerable features unchanged. Existing mesh saliency can not be used in mesh resizing directly because of the neglect of resizing direction. We propose a new mesh saliency that is suitable for mesh resizing. Firstly, we bring up a region descriptor based on its vulnerability to a resizing direction, and use this descriptor to compute the region’s saliency based on its contrast to neighboring regions. We build hierarchical coarse-to-fine segmentations of the input mesh, and evaluate the saliency value on different levels of segmentations. Finally these saliency values are integrated into one saliency map after applying non-linear suppression. Equipped with the saliency map, a framework for non-homogeneous mesh resizing is presented. We regard every edge as a spring, and scale the mesh by stretching the edge. We also add an Laplacian item to the energy function to avoid dramatic shape changes.(4) The previous mesh resizing approaches usually create models with no visible artifacts, while can not precisely preserve the important features in an engineering sense. We present a novel approach for non-homogeneous mesh resizing, which can precisely preserve mesh features. The resizing is achieved by warping a B-spline volumetric grid over the mesh. Control points of B-spline volume are divided into three categories, and processed separately. The control points related to the key features are hard constrained to precisely preserve the features. And the other control points are scaled by a quadratic optimization to distribute the distortion globally.