【作者】 梁艳；

【导师】 章云；

【作者基本信息】 广东工业大学 ， 控制科学与工程， 2017， 博士

【摘要】 人脸识别因其自然、友好、对用户干扰少等优点,成为计算机图像处理领域中极具应用前景的关键技术。经过二十多年的发展,二维人脸识别的研究已取得了很大的进展,在一定约束条件下能取得较好的识别结果。但由于易受光照、姿态、表情等因素的影响,人脸识别技术的发展受到制约。近年来,随着三维数据采集技术的日益成熟,三维人脸识别受到越来越多的关注,涌现了大量有关三维人脸识别算法的研究。本文主要针对非可控条件下的三维人脸识别问题展开深入研究,给出了有效的解决方案。本文的主要研究内容和贡献如下:(1)提出一种对姿态和表情不敏感的三维人脸标志点定位方法。利用HK曲率分析检测候选标志点,根据对面部形状的先验知识,提出一种基于人脸几何结构的分类策略对候选标志点进一步细分,通过把候选标志点与面部标志点模型进行匹配,实现标志点的精确定位。首先在CASIA数据集对该方法的标志点定位精度进行测试,然后在UND/FRGCv2.0数据集对该方法与其他先进的方法进行比较。实验结果表明该方法在姿态和表情变化很大的情况下仍具有高精度和高鲁棒性。(2)提出一种新的表情不变三维人脸识别方法。三维人脸被划分成一组等距测地线条纹,并使用3DWW和质心距离描述条纹的空间关系。此外,提出了相应的相似性度量方法。在CASIA数据库和FRGC v2.0数据库上进行实验。结果表明,即使在表情变化强烈时,该方法仍获得很好的识别性能。(3)针对大幅度姿态变化问题,提出了一种新的姿态不变三维人脸识别方法。利用一个自动的标志点检测器来估计人脸模型的姿态,并使用一个参考模型配准各个人脸模型。利用半脸匹配,该方法可以无缝地处理正面和侧面人脸模型。在Bosphorus和UND/FRGC v2.0数据库中进行的实验表明,该方法对姿态变化具有较高的准确率和鲁棒性。(4)提出一种基于关键点和局部描述符的三维人脸识别算法。首先在尺度空间中检测平均曲率的极值点,作为人脸曲面上的关键点。然后,提出一种网格边缘点过滤器的方法,用于过滤边缘关键点。设计一个由几何形状和形状指数的连接直方图组成的描述符向量来描述关键点的局部形状。最后,使用多任务SRC匹配器实现人脸的识别。在Bosphorus数据库上进行实验,证明了该方法对大的表情变化、姿态变化和遮挡的有效性。更多还原

【Abstract】 Face recognition has been a key technology in the field of computer image processing because it is natural,friendly,and non-disturbing.In the last two decades,2D face recognition research has made great progress.Most 2D face recognition algorithms can achieve reliable recognition performance under controlled conditions.However,the development of 2D face recognition is restricted because it is sensitive to illumination,pose,and expression.With the rapid development of the 3D scanning techniques,3D face recognition has received growing attention in recent years.A large number of studies on 3d face recognition have emerged.In this paper,we mainly study the problem of 3d face recognition under uncontrolled conditions and give the effective solution.The main research content and contributions are as follows:(1)A method for 3D facial landmark localization is presented.The method is insensitive to pose and expression.Candidate landmarks are detected using HK curvature analysis.According to the priori knowledge on facial shape,a facial geometrical structure-based classification strategy is proposed to subdivide the candidate landmarks.Landmark localization is obtained by matching candidate landmarks with a Facial Landmark Model(FLM).The landmark localization accuracy of our method is first experimented on the CASIA dataset.Then,our method is compared with the state-of-the-art methods on the UND/FRGC v2.0 dataset.Experimental results confirm that our method achieves high accuracy and robustness both to large pose and expression variations.(2)A new method for expression-invariant 3D face recognition is proposed.A 3D face is partitioned into a set of iso-geodesic stripes and the spatial relationships of stripes are described by 3D Weighted Walkthrough(3DWW)and centroid distance.Moreover,the way of similarity measure is given.Experiments are performed on the CASIA dataset and the FRGC v2.0 dataset.The results show that the proposed method has advantages on recognition performance despite large expression variations.(3)A new method for pose-invariant 3D face recognition is proposed to handle significant pose variations.It uses an automatic landmark detector to estimate pose for each facial scan.Subsequently,a reference model is registered to the scan.By using the half face matching,it can seamlessly handle frontal and side facial scans.Experiments carried out on the Bosphorus and UND/FRGC v2.0 databases show that the proposed method has high accuracy and robustness to pose variations.(4)A 3D face recognition approach based on keypoints and local descriptors is presented.Firstly,keypoints on the facial scan are detected as mean curvature extrema in scale space.Then,a Mesh Edge Point Filter is proposed to remove edge keypoints.A descriptor vector which consists of concatenated histograms of geometric shapes and shape indices is designed to describe local shapes of keypoints.Finally,the identity of a probe scan is determined by using a multitask SRC,Results from experiments on the Bosphorus database demonstrate the effectiveness of the proposed approach in the presence of large expressions variations,large pose variations,and occlusions.更多还原