一种基于特征融合的人脸识别新方法

A Novel Face Recognition Method Based on Feature Fusion

提出了一种基于特征融合的人脸识别新方法。首先采用两种不同的K L变换分别降低原始图像空间的维数,避开人脸识别小样本集的局限,然后利用复向量将同一样本的两组特征向量合并在一起,通过运用具有统计不相关性的复线性鉴别分析来抽取人脸图像的有效鉴别特征,最后在ORL人脸库上实验结果表明所提出的方法不仅识别性能优于经典的Fisherfaces,而且仅用14个特征识别率就达到96%。

A novel face recognition method based on feature fusion is presented in this paper.First of all,in order to avoid the difficulty of the withinclass scatter matrix being singular,we use two kinds of KL transform methods to respectively reduce the dimensionalities of the original image vector space.Then,the resulting two kinds of features are combined together by a complex vector,and the developed the uncorrelated linear discriminant analysis is employed for feature extraction in the complex feature space.Finally,experimental results on ORL face database show that the proposed method not only is more effective than the classical Fisherfaces,but also achieves a recognition accuracy of 96% using only 14 features.