基于血流图的小波域分块DCT+FLD红外人脸识别方法  

Infrared Face Recognition Method Using Blood Perfusion and Sub-block DCT+FLD in Wavelet Domain

作　　者：谢志华[1,2] 伍世虔[1] 方志军[1] 

机构地区：[1]江西财经大学信息管理学院,南昌330013 [2]江西科技师范学院光电子与通信重点实验室,南昌330013

出　　处：《计算机科学》2009年第12期290-293,共4页Computer Science

基　　金：国家自然科学基金(60665001);江西省教育厅科技项目(GJJD9296)资助

摘　　要：为了从生物特征角度同时结合人脸的局部特征和整体特征提高红外人脸的识别性能,提出了一种基于血流图的小波域分块DCT+FLD(Fisher线性判别)红外人脸识别方法。首先利用血流模型把温谱图转换成血流图,然后用小波变换对人脸血流图像做两级小波分解,再对低频子带进行分块并对每个分块进行DCT变换,提取部分变换后的系数作为子块的特征值,对这些子块的特征值构成的组合特征值从整体上做Fisher线性分析,得到特征子空间,最后根据欧氏距离和三阶近邻分类器进行识别,得到最终的识别结果。实验表明,同基于传统PCA+FLD,DCT+FLD以及DWT+PCA+FLD方法相比,所提出的方法得到了更好的识别效果。To get the good performance of infrared face recognition from the biological feature and Combine the local and whole charaeter,a novel method based on based for infrared face recognition was developed. A new method based on FLD for feature extraction was presented combing blood perfusion and block DCT in wavelet domain was proposed. Firstly, thermal images were converted into blood perfusion domain by blood perfusion model to obtain consistent facial images without effect of ambient variations. Secondly blood perfusion data were decomposed using two scales＇ discrete wavelet transform. Then, the component of low frequency sub-bands was partitioned into sub-blocks, to which the DCT was further applied. The FLD was applied to the global features combined by the extracted coefficients from all sub-blocks in DCT domain. Finally, Euclidean distance and the 3-NN classifier were utilized in recognition. The experiments conducted illustrate that the method proposed in this paper has better performance compared with traditional PCA, PCA ＋FLD in thermal images.

关 键 词：Fisher线性判别(FLD) 血流图 离散余弦变换(DCT) 红外人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TQ639.3[自动化与计算机技术—计算机科学与技术]