基于矩－傅里叶描述子的不同姿态三维人脸识别

摘要

Three-dimensional human face contains richer information compared to 2D face ,but also able to o-vercome gestures,facial expressions and lighting factors,a better representation of facial feature,which gradual-ly widespreads concern and attention.In this paper,firstly,differential geometry theory correction is used in three-dimensional human face depth maps in different poses to the median plane,and then by extracting the contour characteristics of the human face portion,the three -dimensional human face into a two -dimensional curve is easy to handle ,and for how better able to describe the two-dimensional curves,a new method combi-ning the moments and Fourier descriptors is proposed.In the end,face recognition is performed using the extrac-ted curve feature.Experimental results show that this method has good real-time performance and greater ability to distinguish between shapes,while having excellent robustness superior to the traditional Fourier method de-scribed in sub-extracting contour features.%三维人脸相比二维人脸包含更丰富的信息，而且能够克服姿态，表情，光照因素的影响，更好地表示人脸特征，从而逐渐受到广泛的关注和重视。文中对不同姿态下的三维人脸深度图，先用微分几何相关知识把该图校正到正中面，然后通过提取人脸面部的等高线特征，将三维人脸变成容易处理的二维曲线，针对如何能够更好的描述该二维曲线，提出了一种把矩和傅里叶描述子相结合的新方法，最后利用提取的曲线特征进行人脸识别。实验结果表明，该方法具有很好的实时性和更强大的形状区分能力，鲁棒性高，优于传统傅里叶描述子提取等高线特征的方法。