摘要： 网格参数化作为数字几何处理的基本工具，在游戏娱乐、工程设计、仿真模拟等多种领域有着广泛的应用背景。传统的网格参数化方法大多通过求解线性系统或者非线性系统获得结果，存在着求解速度慢、不够鲁棒的问题。提出了一个基于图卷积网格自编码器的网格参数化模型，采用了图卷积网格自编码器的编码部分与自行构建的参数化解码部分结合的方式生成网络，使用一类人脸网格数据集作为网络训练数据，并与传统优化算法进行对比。结果表明，使用建立的网格参数化模型，在保证参数化效果的同时，获得参数化结果的速度比SLIM(Scalable Locally Injective Mappings，SLIM)算法快68%，比PP(Progressive Parameterizations)算法快约4倍。

中图分类号： TP393
文献标识码： A
DOI： 10.19358/j.issn.2096-5133.2020.10.003
引用格式： 高晨. 基于图卷积网格自编码器的网格参数化[J].信息技术与网络安全，2020，39(10)：11-17.

Mesh parameterization based on convolutional mesh autoencoders

Abstract： As a basic tool of digital geometry processing, mesh parameterization has a wide range of applications in game entertainment, engineering design, simulation and other fields. Most of the traditional meshing parameterization methods obtain results by solving linear or nonlinear systems. This article presents a mesh parameterization model based on convolutional mesh autoencoders, the network model is generated by the combination of the encoding part of a convolutional mesh autoencoders and the decoding part, and using a class of human face mesh data set as the network training data. The results show that the algorithm is more than 68 percent faster than the SLIM(Scalable Locally Injective Mappings)algorithm and more than four times faster than the PP(Progressive Parameterizations)algorithm, while the parameters of the mesh parameterization are used to ensure the parameterization effect.

Key words : mesh parameterization；convolutional neural network；autoencoders；convolutional mesh autoencoders