【摘要】 为了在人脸姿态和表情归一化后减少人脸外观的信息损失,提出自适应三维形变模型(3DMM)结合流形分析的人脸识别方法。首先,描述人脸姿态变换引起的2D和3D坐标的不对应性,提出自适应3DMM拟合方法;然后,通过三维变换来保留尽可能多的身份信息,将整个图像网格化映射成3D对象,姿态和表情的归一化保证了变换的稳定;最后,利用多流形判别分析计算流形与流形之间的距离,并利用最近邻分类器完成识别。在Multi-PIE,LFW以及自己采集的数据库上的实验验证了所提方法的有效性,在3个数据库上的识别率分别高达99.8%,95.25%,98.62%。所提方法显著改善了人脸识别性能,在约束和无约束环境下均优于其他几种较新的识别方法。更多还原

【Abstract】 In order to reduce the loss information of face appearance after the normalization of face pose and expression,a normalization face recognition method of face pose and expression based on adaptive three-dimensional morphable model(3DMM)and multiple manifold discriminant analysis was proposed.Firstly,face pose 2D and 3D coordinate transformation caused by the non-correspondence is described,and an adaptive 3DMM fitting method is proposed.Then,the entire image is mapped into a 3Dgrid objects by three-dimensional transformation to preserve the identity information as much as possible.Finally,multiple manifold discriminant analysis is used to calculate the distance between manifolds,and the nearest neighbor classifier is used to finish recognition.The effectiveness of the proposed method is verified by experimental results on data base Multi-PIE,LFW and self-collection experiments,the face recognition accuracy on the three databases can achieve 99.8%,95.25%,98.62%,respectively.The proposed method significantly improves the performance of face recognition,and it is better than other similar advanced methods in constrainted and unconstrained environment.更多还原