Two approaches are proposed for cross-pose face recognition, one is based onthe 3D reconstruction of facial components and the other is based on the deepConvolutional Neural Network (CNN). Unlike most 3D approaches that considerholistic faces, the proposed approach considers 3D facial components. Itsegments a 2D gallery face into components, reconstructs the 3D surface foreach component, and recognizes a probe face by component features. Thesegmentation is based on the landmarks located by a hierarchical algorithm thatcombines the Faster R-CNN for face detection and the Reduced Tree StructuredModel for landmark localization. The core part of the CNN-based approach is arevised VGG network. We study the performances with different settings on thetraining set, including the synthesized data from 3D reconstruction, thereal-life data from an in-the-wild database, and both types of data combined.We investigate the performances of the network when it is employed as aclassifier or designed as a feature extractor. The two recognition approachesand the fast landmark localization are evaluated in extensive experiments, andcompared to stateof-the-art methods to demonstrate their efficacy.
展开▼
机译:提出了两种用于跨姿势人脸识别的方法,一种是基于人脸组件的3D重建,另一种是基于DeepConvolutional神经网络(CNN)。与大多数考虑整体面孔的3D方法不同,所提出的方法考虑3D面部组件。将2D画廊面细分为组件,为每个组件重建3D曲面,并通过组件特征识别探针面。分层是基于分层算法定位的地标,该算法将用于面部检测的Faster R-CNN和用于地标定位的Reduced Tree StructuredModel相结合。基于CNN的方法的核心部分是可见的VGG网络。我们研究了在训练集上具有不同设置的性能,包括3D重建的合成数据,来自野生数据库的真实数据以及两种类型的数据组合。我们研究了使用网络时的性能作为分类器或设计为特征提取器。在广泛的实验中对这两种识别方法和快速地标定位进行了评估,并与最新方法进行了比较以证明其有效性。
展开▼