Abstract: To get the good performance of infrared face recognition from the biological feature and statistical character, this paper proposes a novel method for infrared face recognition based on　blood perfusion image by combining discrete cosine transform (DCT) and features selection. Firstly, infrared thermal images are converted into blood perfusion domain by blood perfusion model to get the constant information. Secondly, DCT is chosen to reduce the correlation in original face image. Finally, to improve the effectiveness of features extraction in DCT domain, the objectives of features selection and subspace learning are consistent (both follow the separability discriminant criterion): a feature selection algorithm is proposed to extract the DCT coefficients, LDA (linear discriminant analysis) is applied to DCT coefficients extracted by the feature selection algorithm. The experimental results illustrate that the proposed method can quickly and efficiently extract the features of blood perfusion domain for classification, and get better recognition performance than traditional DCT+LDA method.

Key words: infrared face recognition, blood perfusion image, discrete cosine transform (DCT), separability, linear discriminant analysis (LDA)

摘要： 为了从生物特征和统计角度来提高识别的性能，提出了一种基于血流图的离散余弦变换（discrete cosine transform，DCT）与特征选择相结合的人脸识别方法。该方法首先利用血流模型把红外温谱图转换成血流图，得到更具丰富频率的特征。其次，采用DCT变换可以有效地消除血流图的相关性。最后，在DCT域特征提取阶段，为了提高特征提取的有效性，特征选择和子空间学习基于一致的可分性目标：特征选择引入基于可分性的DCT系数选择算法以抽取鉴别能力强的DCT系数，对抽取的DCT系数采用基于可分性的线性鉴别分析（linear discriminant analysis，LDA）方法。实验结果表明，该红外人脸识别方法可以快速有效地提取血流图中适合分类的特征，识别率优于传统DCT+LDA方法。

关键词: 红外人脸识别, 血流图, 离散余弦变换（DCT）, 可分性, 线性鉴别分析（LDA）