基于三维模型的多姿态人脸识别

Pose Invariant Face Recognition Based on 3D Model

提出了一种利用三维人脸模型匹配二维人脸图像的分层人脸识别方法和基于模糊数学的人脸姿态角度估计算法。对多姿态二维图像进行姿态空间划分,利用主成分分析方法(PCA)形成多姿态特征脸。识别过程首先估计测试图像姿态和模糊姿态角,在估计的姿态空间内采用基于PCA的方法进行第一层识别得到候选个体,然后利用候选个体的三维模型结合模糊姿态角产生虚拟图像,利用相关进行第二层识别。实验结果表明,该方法对姿态的变化有较好的鲁棒性。

An estimating pose angle method based on fuzzy math and a layered face recognition method by matching 2D probe image to 3D model were proposed in this paper. After classifying multi-pose images to different pose space, Primary Component Analysis （PCA） was used to get eigenfaces in the given pose space. While a probe image was recognized, its pose and fuzzy angle were estimated firstly and then they were matched in the estimated pose space by PCA method. Some candidates were gotten by upwards step and their 3D model were employed to generate dynamically virtual 2D images with view angles nearby fuzzy angle. Image correlation was employed as a classifier to match the probe image to the virtual images. The experiment result shows that the proposed method is robust to pose variant.