Abstract: In recent years, more attention has been paid to infrared face recognition since infrared face images have lots of special properties such as defense from camouflage, cheat and independence to the ambient light. A novel method for infrared face recognition based on Gabor Wavelet and Singular Value Decomposition（SVD） is proposed. The normalized infrared face image is first decomposed by convolving with multi-scale and multi-orientation Gabor filters, obtaining many Gabor feature matrixes. SVD is performed on every Gabor feature matrix and then the largest singular values of every matrix are combined to form the final infrared face feature vector. Finally, a radial basis function neural network is used for classification. The experimental results on infrared face database show that, compared to traditional identification methods, this method has good recognition effect.

Key words: Gabor wavelet, Singular Value Decomposing（SVD）, RBF neural networks, infrared face recognition

摘要： 由于热红外人脸图像具有防伪装、防欺诈以及独立于环境光照的特点，所以近年来热红外人脸识别问题备受关注。提出一种基于Gabor小波和SVD的热红外人脸识别新方法。对归一化后的热红外人脸图像进行多方向多尺度Gabor变换，得到多个Gabor特征矩阵；对每个矩阵进行奇异值分解，并把每个矩阵最大的奇异值组合起来作为最终的热红外人脸特征向量；使用径向基神经网络进行分类识别。在自建热红外人脸数据库上的实验结果表明，相比于传统的识别方法，该方法具有较好的识别效果。

关键词: Gabor小波, 奇异值分解（SVD）, RBF神经网络, 热红外人脸识别