词汇树融合SIFT的热成像和视觉图像人脸识别  

Fusion of Vocabulary Tree and SIFT for Face Recognition in Thermal Image and Visual Image

作　　者：张国平[1] 周改云[1] 马丽[1] 

机构地区：[1]平顶山学院软件学院,河南平顶山467000

出　　处：《电视技术》2014年第23期158-163,共6页Video Engineering

基　　金：国家自然科学基金项目(U1204611);河南省科技厅科技发展计划项目(134300510037);平顶山学院青年科研基金项目(PXY-QNJJ2013010)

摘　　要：针对热成像和视觉图像人脸识别问题,提出了一种基于词汇树融合尺度不变特征变换方法。首先,对视觉和热成像图像分别单独进行提取,利用Viola-Jones层叠检测器从自然图像中检测出人脸;然后,利用SIFT描述符从尺度空间提取稳定特征;最后,使用词汇树进行分类,利用评分融合和决策融合算法提高系统的精确性和安全性。在拍摄的41个人的脸部图像上的实验表明了该方法的有效性,识别率可接近100%,相比其他几种较为新颖的人脸识别方法,该方法取得了更高的识别精度,并且在一定程度上降低了计算耗时。For the face recognition problem from thermal image and visual image, a fusion method based on vocabulary tree and scale invariant feature transform is proposed. Firstly, thermal image and visual image is exteacted respectively, Viola -Jones cascading detector is used to detect face from naturM images. Then, SIFF descriptors are used to extract stabihty features from invariant space. Finally, classification is finished by vocabulary tree, score fusion and decision fusion algorithm is used to improve the accuracy and security of system. The effectiveness of proposed method has been verified by experiments on face images of 41 person gathered self, recognition accuracy of proposed method can achieve 100%, experimental results show that proposed mehtod has higher recognition accuracy and lower computing time than several other advanced algorithms.

关 键 词：词汇树 尺度不变特征变换 热成像 视觉图像 人脸识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]