摘要
针对点云数据的三维重建问题,提出了一种隐曲面重构的广义多项式神经网络新方法。该广义多项式神经网络隐层各神经元激励函数互不相同且线性无关,能够对应地学习点云数据样本中不同的模式,因此,具有较好的学习能力。基于梯度下降法原理,推导了其学习算法。仿真实验尝试将该方法应用于一些简单封闭物体的带噪点云数据隐式曲面重建,取得了较理想的重建质量和去噪效果。
A new type of generalized polynomials neural network was proposed to reconstruct 3D implicit surface from the scattered points. Since its hidden-layer neurons were activated with the different and linear independent generalized polynomials, the proposed neural network could achieve good performance in learning different patterns. Then the weightsupdating formula for the new type of neural network was derived based on gradient-descent method. The simulation results on some scattered point models show that this method can obtain good reconstruction quality and denoising effect.
出处
《计算机应用》
CSCD
北大核心
2009年第8期2043-2045,2064,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(60775050)
浙江大学CAD/CG国家重点实验室开放课题
关键词
广义多项式
神经网络
隐式曲面
点云
generalized polynomial
neural network
implicit surface
scattered point