应用于小样本的差异字典人脸识别

针对人脸识别中每个人只有小规模训练样本的情况,在基于表示的分类(representation based classification,RBC)方法基础上使用由无关类组成的差异字典。差异字典一般由具有面部姿态变化与表情变化的人脸及其基准脸构成,需要训练样本为基准脸才能得到较好的识别效果。为防止小规模训练样本中有非基准脸使差异字典出现识别效果下降的情况,使用灰度对称脸将训练样本中的非基准脸转换为近似基准脸,进行差异字典的训练。实验结果表明,该人脸识别方法在小样本情况下的ORL、GT(Georgia tech)、FERET人脸库上具有良好的表现。

Difference dictionary face recognition applied to small samples

Aiming at the face recognition with only small-scale training samples per person,the difference dictionary consisting of unrelated-class possible facial gestures and expression changes and their reference faces was used,based on the representation-based classification(RBC)method.However,the use of the difference dictionary required the training sample as the reference face to get better recognition effects.To prevent the non-reference face in the small-scale training sample from degrading the difference dictionary recognition effects,it was proposed to convert the non-reference face in the training sample into the approximate reference face by using the grayscale symmetrical face.Experimental results show that the face recognition method has good performance on ORL,GT(Georgia tech)and FERET face database in small sample cases.