自适应通用学习框架在人脸识别中的应用研究

ON APPLYING ADAPTIVE GENERIC LEARNING FRAMEWORK TO FACE RECOGNITION

现实生活中,人脸识别系统通常必须面对单样本每人（SSPP）的问题,即在数据库中每个人只有1张训练样本。这种情况下,系统不能很好地学习训练样本的判别信息,因而许多流行的人脸识别方法将不能很好地奏效。为了解决这个问题,自适应通用学习（AGL）方法利用一个通用判别模型来更好地区分各个单训练样本,同时,采用双线性表示算法来推测类间矩与类内矩;使得FLDA可以应用于单样本人脸识别。在ORL及FERET的实验表明,与其他几种常用的方法相比较,AGL在处理单样本人脸识别问题上取得了更好的结果。

In real lives, face recognition systems usually have to encounter single sample per person （SSPP） problem, that is, only one training sample is enrolled for one person in database. In such case, the system cannot well learn the discriminant information of training sam- ple, so that many popular face recognition methods fail to work well. To address this problem, the adaptive generic learning （AGL） method uses a generic discriminant model to better distinguish every single training sample, meanwhile it adopts the bilinear representation algorithm to infer the inter-class moment and the intra-class moment, which makes Fisher＇s linear discriminant analysis （FLDA） be able to be applied to single sample face recognition. Experiments on ORL and FERET face database show that the AGL method achieves better result on SSPP problem compared with other common solutions.