基于得分融合近红外与可见光图像的人脸识别.docVIP

基于得分融合近红外与可见光图像的人脸识 别 潘磊1，尹义龙1，李徐周1,2 1 山东大学计算机科学与技术学院学院，济南　250101 2 山东青年政治学院学院，济南　250103 摘要： 基于人脸的身份鉴别技术已受到越来越多的重视。在可见光环境里，人脸识别算法对 光照的变化非常敏感。使用近红外人脸图像进行识别可以有效的提高人脸识别对光照的鲁棒 性，但是近红外图像成像的原理会导致很多面部特征的丢失。所以融合近红外人脸图像和可见 光人脸图像既可以提高人脸识别算法对光照的鲁棒性，又能保证算法的性能。据此提出一种基 于匹配得分融合近红外图像和可见光图像的人脸识别算法。首先使用特征脸（Eigenfaces）算 法获得两种图像各自的匹配得分，然后进行基于得分的融合获得最终得分。在CASIA HFB人 脸数据库和HITSZ Lab1人脸数据库上进行实验,实验结果证明了此方法的识别效果优于各单一 条件下的识别性能。 关键词：模式识别；人脸识别；近红外；可见光；得分融合；特征脸 中图分类号： TP391 Face Recognition Based on Score-level Fusion of Near Infrared and Visible Light Images Pan Lei1, Yin Yilong1, Li Xuzhou1,2 1 School of Computer Science and Technology, Shandong University, Jinan 250101 2 Shandong Youth University of Political Science, Jinan 250103 Abstract: More attentions have been paid to face authentication recently. The performance of face recognition algorithms is sensitive to illumination in visible light environment. Face recognition algorithms using the near infrared face image are robust against illumination. But the principle of near infrared imaging brings the loss of facial features. Fusing the near infrared and visible light face images for recognition not only can improve the robustness of face recognition algorithms against illumination, but also can keep the performance of the algorithms. A new method based on score-level fusion of the near infrared and the visible light face images is proposed. Firstly, matching scores of near infrared and visible light face images are got by Eigenfaces separately. Second is to fuse the scores to get the ?nal score. The method are veri?ed on CASIA HFB face database and HITSZ Lab1 face database. The results shows the e?ectiveness of the proposed method. 基金项目： 高等学校博士学科点专项科研基金项目(20100131110021)，山东省高等学校科技计划项目(J11LG28) 作者简介： 潘磊，男，硕士，主要研究方向：模式识别，人脸识别。通信作者：尹义龙，男，教授，主要研究方向：机器学习 与数据挖掘，生物特征识别。 -1- Key words: Pattern Recognition; Face Recognition; Near Infrared Image; visible Image;Score-level Fusion; Eigenfaces  0  引言  人脸识别是利用计算机获取人脸图像并进行分析预处理，然后以特定方法提取