基于因素分解模型的两步人脸识别  

Two-step Face Recognition Based on Factorization Models

作　　者：程载和[1] 

机构地区：[1]无锡职业技术学院,无锡214121

出　　处：《计算机科学》2017年第B11期263-266,共4页Computer Science

摘　　要：为了减轻人脸识别中表情以及姿态等因素变化对识别结果的影响,Xu提出了利用原始样本和对称样本的两步人脸识别算法。但当人脸图像受外在因素干扰产生较大变化时,该方法的识别结果并不理想。因此提出了一种基于因素分解模型的两步人脸识别算法。新算法在特征提取过程中利用因素分解模型将"身份因素"和"表情因素"从人脸图像中分离出来,加以控制。然后提取测试集图像中的新身份和新表情,并将其与训练集中的旧身份或旧表情相互作用,合成新的人脸图像。同时为了保证分类精度,在识别阶段针对原始样本和合成样本分别采用两步人脸识别的方法,充分利用了分数层次融合的优势,进一步提高了算法的识别效果。To overcome the effect of facial expression and pose on face recognition,Xu proposed using the original and 'symmetrical face' training samples to perform representation based two-step face recognition.However,it usually gets bad symmetrical face samples based on mirror image with the changes of facial poses,which may affect the accuracy of recognition.A two-step face recognition algorithm based on factor decomposition model was proposed.Firstly,to reduce the effect of light,pose and other factors,the factorization mode[is used to separate the 'expression factors' and 'identity factor' from the face images which can be controlled,to construct new image samples.Meanwhile,the two-step face recognition method is adopted in the recognition stage,which makes full use of the advantage of the score level fusion,and further improves the recognition rate of the proposed algorithm.

关 键 词：人脸识别 表情因素 因素分解模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]