摘要
3D人脸在采样不完整时,很难对其标定特征点,进而直接影响识别正确率。本文针对这个问题提出了一种基于局部曲线特征的人脸识别方法。该方法首先计算了3D人脸在深度空间,曲率空间,和测地空间下的表示,采用SURF算法自动提取特征点,通过神经网络的方法分析特征点之间的连接线上的深度,曲率和测地信息来达到识别的目的。实验结果表明,这种自动识别特征点的方法相对于基于标定的识别方法具有更强的鲁棒性和更高的正确率。
It＇s diffieuh to find feature points in 3D facial meshes when there are missing parts on these face scans. This wfwk proposes a new reeognilion method based on hwal feature rurve. Firstly, we calctdale the projertions of 3D fares in depth, cufvahtre and geodesic eoordinale system, then apply SURF on these projections Io find out teature points, finally, we use artificial neural network to analyze the curve between feature points and thus face scans arc recognized. The resuh shows that Ihe robustness and acctwaey of our method is higher than the method hased on hmdmark.
出处
《科技视界》
2016年第3期128-129,共2页
Science & Technology Vision
