基于混合频谱信号编码的网格纹理平滑  

Mesh Texture Smoothing Based on Hybrid Spectral Encoding

作　　者：郭艺辉 陆寄远 黄承慧 钟雪灵 林淑金[2] 苏卓[3] 罗笑南 GUO Yi-Hui;LU Ji-Yuan;HUANG Cheng-Hui;ZHONG Xue-Ling;LIN Shu-Jin;SU Zhuo;LUO Xiao-Nan(School of Internet Finance and Information Engineering,Guangdong University of Finance,Guangzhou 510521;School of Communication and Design,Sun Yat-sen University,Guangzhou 510006;School of Computer Science and Engineering,Sun Yat-sen University,Guangzhou 510006;School of Computer Science and Information Security,Guilin University of Electronic Technology,Guilin,Guangxi 541004)

机构地区：[1]广东金融学院互联网金融与信息工程学院,广州510521 [2]中山大学传播与设计学院,广州510006 [3]中山大学计算机学院,广州510006 [4]桂林电子科技大学计算机与信息安全学院,广西桂林541004

出　　处：《计算机学报》2021年第2期318-333,共16页Chinese Journal of Computers

基　　金：国家自然科学基金项目(71501051,61872394,61772149);广东省普通高校人文社会科学研究重点项目(2018WZDXM032);广东省普通高校科研平台和科研创新项目(2020KTSCX085);广东省自然科学基金(2017A050501042);广州市科技计划项目(202002030473);广东省基础与应用基础研究基金(2019A1515011953)资助.

摘　　要：网格纹理平滑技术要求既能保持模型大尺度结构特征又能去除模型小尺度纹理.然而当模型小尺度纹理与噪声相差较大时,大多数网格光顺算法会将网格纹理识别为特征加以保持,而无法有效将其去除;现有的基于谱分析的网格光顺方法尽管能有效去除网格纹理,但又无法同时保持模型大尺度结构特征.为解决该问题,本文提出一种基于混合频谱信号编码的低通过滤网格纹理平滑算法.首先采用基于视觉感知的特征识别方法,准确区分模型大尺度与小尺度特征.然后,基于顶点特征尺度,采用差异性频谱信号编码的方式进行几何信息重建,最终实现在保持网格模型大尺度结构特征的同时有效去除小尺度纹理.算法解决了现有网格光顺方法在模型小尺度纹理特征与噪声有明显区别时,无法有效去除纹理的问题;并且也解决了现有基于谱分析的网格光顺方法无法在去除模型小尺度纹理的同时,又能保持模型大尺度特征的矛盾.实验结果验证了算法的有效性.More and more application systems,such as mesh model reuse,3D texture mapping,3D data transmission,mesh compression,simplification,3D real-time rendering and so on,have put forward requirements for the 3D mesh textures smoothing.The technology of the mesh texture smoothing is expected to both reduce the small-scale detail texture features and keep the large-scale intrinsic structures.Traditional mesh smoothing methods tend to focus on removing high frequency random noise and preserving the features.In case of the small-scale textures are quite different from noise,those methods tend to regard them as features to preserve them rather than eliminate them.The existing mesh smoothing methods based on spectral analysis can smooth out all of the small-scale textures,but also over-smooth the large-scale structural features on the models.To solve these problems,the paper proposed a low-pass filter based on the hybrid spectral encoding.Firstly,a feature recognition method based on the visual awareness is used to accurately recognize the scale-features on the models.The mesh Laplace-Beltrami operator is constructed and the base functions are obtained through the spectral analysis.Regarding the geometric informations of the vertices as signals,a spectral space is constructed by projecting the geometric informations to the base functions.Using the low-frequency coefficients,a smooth base surface of the original mesh model is constructed,which is regarded as the three-dimensional datum of the original mesh model.The height between the mesh vertex and the three-dimensional datum is calculated to obtain the visual importance of the vertex.The vertex with the height value larger than a threshold is defined as the large-scale feature vertex.Next,a hybrid spectral encoding method is proposed to reconstruct the mesh model.There are two frequencies setted appropriately:one is the higher frequency β which is used to remove high frequency noise and construct structural features,and the other is the lower frequency α which is used to

关 键 词：网格纹理平滑 网格光顺 模型尺度特征 视觉感知 谱图理论 混合频谱编码 数字几何处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]