基于核稀疏表示的人脸人耳融合识别算法的研究
|
摘要
针对人脸人耳融合识别算法对图像光照变化、表情变化、拍摄角度变化等鲁棒性不强的问题,将核稀疏表示理论引入到人脸人耳融合识别中,提出基于核稀疏表示的人脸人耳融合识别算法。新算法采用的是能有效降低样本维度的PCA特征提取算法,人脸人耳的特征融合层级选用既能实现冗余信息有效压缩,又能最大程度利用不同模态生物特征可区分性的特征级融合。考虑到不同模态生物特征对最终识别的贡献可能有所不同,该算法采用加权串联融合法,同时测试样本在训练样本中稀疏表示系数的求解采用的是迭代速度比较快的正交匹配追踪算法。与其他识别算法相比,该算法具有非常好的识别性能,并且对人脸人耳图像变化具有很强的鲁棒性。
Since the human face and ear fusion recognition algorithm has weak robustness in variations of image illumination,facial expression and shooting angle,the kernel-sparse representation theory is introduced into the human face and ear fusion recognition,and a human face and ear fusion recognition algorithm based on kernel-sparse representation is proposed. The PCA feature extraction algorithm that can effectively reduce sample dimensions is adopted in the algorithm. The feature-level fusion that can not only realize effective compression of redundant information,but also make the best use of the distinguishability of biological features in different modes is selected for the feature fusion level of human faces and ears. The weighted series fusion method is used for the algorithm considering that different modes of biological features may have different contributions to the final recognition. The orthogonal matching pursuit algorithm with fast iteration speed is used to solve the sparse representation coefficient of tested samples in training samples. In comparison with other recognition algorithms,the algorithm has a much better recognition performance,and stronger robustness in variations of human face and ear images.
Since the human face and ear fusion recognition algorithm has weak robustness in variations of image illumination,facial expression and shooting angle,the kernel-sparse representation theory is introduced into the human face and ear fusion recognition,and a human face and ear fusion recognition algorithm based on kernel-sparse representation is proposed. The PCA feature extraction algorithm that can effectively reduce sample dimensions is adopted in the algorithm. The feature-level fusion that can not only realize effective compression of redundant information,but also make the best use of the distinguishability of biological features in different modes is selected for the feature fusion level of human faces and ears. The weighted series fusion method is used for the algorithm considering that different modes of biological features may have different contributions to the final recognition. The orthogonal matching pursuit algorithm with fast iteration speed is used to solve the sparse representation coefficient of tested samples in training samples. In comparison with other recognition algorithms,the algorithm has a much better recognition performance,and stronger robustness in variations of human face and ear images.
引文
[1] SCHEIRER W J,WILBER M J,ECKMANN M,et al. Goodrecognition is non?metric[J]. Pattern recognition,2015,47(8):2721?2731.
[2]崔言伟.基于人耳和侧面人脸的身份识别[D].北京:北方工业大学,2013.CUI Yanwei. Human identification based on the ear and profileface[D]. Beijing:North China University of Technology,2013.
[3]李懿,刘晓东.一种新的模糊稀疏表示人脸识别算法[J].大连理工大学学报,2017,57(2):189?194.LI YI,LIU Xiaodong. A new fuzzy sparse representation algo?rithm for face recognition[J]. Journal of Dalian University ofTechnology,2017,57(2):189?194.
[4]田莹,张德斌,马浩迪.一种稀疏表示的多姿态人耳识别方法[J].系统仿真学报,2014,26(9):2126?2129.TIAN Ying,ZHANG Debin,MA Haodi. Muli?pose ear recogni?tion using sparse[J]. Journal of system simulation,2014,26(9):2126?2129.
[5]袁立,李文明,穆志纯.人脸人耳多模态生物特征模板保护方法研究[J].仪器仪表学报,2012,33(12):2767?2773.YUAN Li,LI Wenming,MU Zhichun. Face and ear multimod?al biometric template protection[J]. Chinese journal of scientificinstrument,2012,33(12):2767?2773.
[6] AMIRTHALINGAM G. Face and ear biometric recognition us?ing PCA and LLE algorithm[J]. International journal of infor?mation science and intelligent system,2014,3(2):47?54.
[7] YANG M,ZHANG L. Gabor feature based sparse representa?tion for face recognition with Gabor occlusion dictionary[C]//Proceedings of 11th European Conference on Computer Vision.Berlin:Springer,2010:448?461.
[8] BENGIO Y,DELALLEAU O,ROUX N L,et al. Learning ei?genfunctions links spectral embedding and kernel PCA[J]. Neu?ral computation,2014,16(10):2197?2219.
[9]王科俊,阎涛,吕卓纹.基于耦合度量学习的特征级融合方法及在步态识别中的应用[J].东南大学学报(自然科学版),2013,43(z1):7?11.WANG Kejun,YAN Tao,LüZhuowen. Feature level fusionmethod based on the coupled metric learning and its applica?tion in gait recognition[J]. Journal of Southeast University(Nat?ural science edition),2013,43(S1):7?11.
[10] ZHANG B,YANG H,YIN Z. A region?based normalizedcross correlation algorithm for the vision?based positioning ofelongated IC chips[J]. IEEE transactions on semiconductormanufacturing,2015,28(3):345?352.
[11]孙子璇,易荣华.基于小波变换的正交匹配追踪算法及其应用[J].计算机科学,2012,39(z3):273?275.SUN Zixuan,YI Ronghua. wavelet transform?based orthorgo?nal matching pursuit algorithm and its application[J]. Computerscience,2012,39(S3):273?275.
[12] SARMADI H,KARAMODIN A,ENTEZAMI A. A new itera?tive model updating technique based on least squares minimalresidual method using measured modal data[J]. Applied mathe?matical modelling,2016,40(23):10323?10341.
[2]崔言伟.基于人耳和侧面人脸的身份识别[D].北京:北方工业大学,2013.CUI Yanwei. Human identification based on the ear and profileface[D]. Beijing:North China University of Technology,2013.
[3]李懿,刘晓东.一种新的模糊稀疏表示人脸识别算法[J].大连理工大学学报,2017,57(2):189?194.LI YI,LIU Xiaodong. A new fuzzy sparse representation algo?rithm for face recognition[J]. Journal of Dalian University ofTechnology,2017,57(2):189?194.
[4]田莹,张德斌,马浩迪.一种稀疏表示的多姿态人耳识别方法[J].系统仿真学报,2014,26(9):2126?2129.TIAN Ying,ZHANG Debin,MA Haodi. Muli?pose ear recogni?tion using sparse[J]. Journal of system simulation,2014,26(9):2126?2129.
[5]袁立,李文明,穆志纯.人脸人耳多模态生物特征模板保护方法研究[J].仪器仪表学报,2012,33(12):2767?2773.YUAN Li,LI Wenming,MU Zhichun. Face and ear multimod?al biometric template protection[J]. Chinese journal of scientificinstrument,2012,33(12):2767?2773.
[6] AMIRTHALINGAM G. Face and ear biometric recognition us?ing PCA and LLE algorithm[J]. International journal of infor?mation science and intelligent system,2014,3(2):47?54.
[7] YANG M,ZHANG L. Gabor feature based sparse representa?tion for face recognition with Gabor occlusion dictionary[C]//Proceedings of 11th European Conference on Computer Vision.Berlin:Springer,2010:448?461.
[8] BENGIO Y,DELALLEAU O,ROUX N L,et al. Learning ei?genfunctions links spectral embedding and kernel PCA[J]. Neu?ral computation,2014,16(10):2197?2219.
[9]王科俊,阎涛,吕卓纹.基于耦合度量学习的特征级融合方法及在步态识别中的应用[J].东南大学学报(自然科学版),2013,43(z1):7?11.WANG Kejun,YAN Tao,LüZhuowen. Feature level fusionmethod based on the coupled metric learning and its applica?tion in gait recognition[J]. Journal of Southeast University(Nat?ural science edition),2013,43(S1):7?11.
[10] ZHANG B,YANG H,YIN Z. A region?based normalizedcross correlation algorithm for the vision?based positioning ofelongated IC chips[J]. IEEE transactions on semiconductormanufacturing,2015,28(3):345?352.
[11]孙子璇,易荣华.基于小波变换的正交匹配追踪算法及其应用[J].计算机科学,2012,39(z3):273?275.SUN Zixuan,YI Ronghua. wavelet transform?based orthorgo?nal matching pursuit algorithm and its application[J]. Computerscience,2012,39(S3):273?275.
[12] SARMADI H,KARAMODIN A,ENTEZAMI A. A new itera?tive model updating technique based on least squares minimalresidual method using measured modal data[J]. Applied mathe?matical modelling,2016,40(23):10323?10341.