一种新的基于单视图的多姿态人脸识别方法

摘要

姿态变化和单视图是二维人脸识别研究的瓶颈问题.本文基于姿态矫正的思想,提出了一种基于单视图的多姿态人脸识别方法.首先,通过多视角主动表观模型进行人脸对齐和归一化;其次,基于线性回归算法寻求正、侧人脸之间的关系,并利用此关系进行姿态矫正得到正脸图像;最后,采用遗传算法筛选支持向量机的参数,并利用支持向量机对矫正后的人脸进行分类.在CAS-PEAL-R1人脸数据库上的实验结果表明,该方法在处理姿态变化的人脸识别问题时,对于姿态为15°、30°和45°的识别率分别达到了98％、84％和76％,识别性能高于其它方法.%The pose variation and single view is a bottleneck problem for recognition of two-dimensional face.A novel method for recognition of pose-invariant face with single image based on the pose correction is proposed in this paper.Firstly,facial feature points are located based on the view-based AAM (Active Appearance Model) and face images are aligned and normalized.Secondly,mapping from the non-frontal image to the frontal image is constructed based on the algorithm for linear regression and frontal faces are obtained from non-frontal faces with different poses.Finally,the SVM (Support Vector Machine) is used to classify the facial features and the parameters of SVM is determined by the genetic algorithm.Experimental results based on the CAS-PEAL-R1 face database show that performance of the proposed approach is better than those by other approaches for pose-invariant face recognition.The recognition rates for face images with pose of 15 °,30 ° and 45 °can reach 98％,84％ and 76％ respectively.