中图分类号：TP391         文献标识码：A         文章编号：2096-4706（2020）21-0071-05

Sketch Face Recognition Based on Residual Network and Multi-mode Triplet Loss

LAN Ling¹，WU Jianbin2，HOU Liang³

（1.Guangdong Beijiang Middle School，Shaoguan 512026，China；2.Education Bureau of Wujiang District，Shaoguan City，Shaoguan 512029，China；3.Shaoguan Public Security Bureau，Shaoguan 512029，China）

Abstract：Face sketch recognition is to recognize face photos from a large face sketch data set，and its main challenge lies in the differences between different modes. In order to solve this problem，a sketch face recognition framework based on multi-task metric learning of residual network is proposed. First，for the problem of reducing the feature difference between different modes，the threechannel neural network is designed to extract the nonlinear characteristics of the photo mode and the sketch mode，and then the parameters of the three networks are shared. Secondly，a multi-mode Triplet Loss is designed to constrain the features in the public space，so that the model expands the distance of heterogeneous samples while reducing similar differences in sketch faces.

Keywords：deep learning；residual network；sketch face recognition；multi-mode Triplet Loss

参考文献：

[1] TANG X O，WANG X G. Face sketch recognition [J].IEEETransactions on Circuits and Systems for Video Technology，2004，14（1）：50-57.

[2] TANG X O，WANG X G. Face Photo-Sketch synthesis andRecognition [J].IEEE Transactions on Pattern Analysis and MachineIntelligence，2009，31（11）：1955-1967.

[3] SAMMA H，SUANDI S A，MOHAMAD-SALEH J.Face sketchrecognition using a hybrid optimization model [J].NeuralComputing and Applications，2019，31（10）：6493-6508.

[4] LIU D C，LI J，WANG N N，et al. Composite componentsbasedface sketch recognition [J].Neurocomputing，2018，302：46-54.

[5] ZHANG W，WANG X G，TANG X O. Coupled informationtheoreticencoding for face photo-sketch recognition [C]//Proceedings ofthe 2011 IEEE Conference on Computer Vision and Pattern Recognition.Washington：IEEE Computer Society，2011：513-520.

[6] KLARE B F，LI Z F，JAIN A K. Matching Forensic Sketchesto Mug Shot Photos [J].IEEE transactions on pattern analysis andmachine intelligence，2011，33（3）：639-646.

[7] GALOOGAHI H K，SIM T. Inter-modality Face Sketch Recognition [C]//Proceedings of the 2012 IEEE International Conferenceon Multimedia and Expo.Washington：IEEE Computer Society，2012：224-229.

[8] ALEX A T，ASARI V K，MATHEW A. Local Difference ofGaussian Binary Pattern：Robust Features for Face Sketch Recognition [C]//Proceedings of the 2013 IEEE International Conference onSystems，Man，and Cybernetics.Washington：IEEE Computer Society2013：1211-1216.

[9] KLARE B F，JAIN A K. Heterogeneous face recognition using kernel prototype similarities [J].IEEE transactions on pattern analysis and machine intelligence，2013，35（6）：1410-1422.

[10] TANG X O，WANG X G. Face Photo-Sketch synthesis and Recognition [C]//Proceedings Ninth IEEE International Conference on Computer Vision.IEEE，2003：687-694.

[11] LIU Q S，TANG X O，JIN H L，et al. A Nonlinear Approach for Face Sketch Synthesis and Recognition [C]//Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Washington：IEEE Computer Society2005：1005-1010.

[12] LI J，YU X Y，PENG C L，et al. Adaptive representationbased face sketch-photo synthesis [J].Neurocomputing，2017，269（20）：152-159.

[13] WANG N N，GAO X B，LI J. Random sampling for fastfacesketch synthesis [J].Pattern Recognit，2018，76：215-227.

[14] ZHANG L L，LIN L，WU X，et al. End-to-endphotosketchgeneration via fully convolutional representationlearning [C]//Proceedings of the 5th ACM on International Conference on MultimediaRetrieval.New York：Association for Computing Machinery，2015：627-634.

[15] WANG N N，ZHA W J，LI J，et al. Back projection：An effective postprocessing method for GAN-based face sketch synthesis [J].Pattern Recognition Letters，2018，107（1）：59-65.

[16] JIAO L C，ZHANG S B，LI L L，et al. A modified convolutional neural network for face sketch synthesis [J].Pattern Recognition，2018，76：125-136.

[17] JIANG J J，YU Y，WANG Z，et al. Graph-Regularized Locality-Constrained Joint Dictionary and Residual Learning for Face Sketch Synthesis [J].IEEE Transactions on Image Processing，2019，28（2）：628-641.

[18] LIN D H，TANG X O. Inter-modality face recognition [C]//ECCV 2006：Computer Vision-ECCV.Springer，2006：13-26.

[19] YI D，LIU R，CHU R F，et al. Face matching between near infrared and visible lightimages.International [C]//Proceedings of the 2007 international conference on Advances in Biometrics.Springer-Verlag，2017：523-530.

[20] SHARMA A，JACOBS D W. Bypassing synthesis：PLS forfacerecognition with pose low-resolution and sketch [C]//Conference onComputer Vision and Pattern Recognition（CVPR）.IEEE，2011：593-600.

[21] MEINA K，SHAN S G，ZHANG H H，et al. Multi-view discriminant analysis [J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2016，38（1）：188-194.

[22] MIGNON A，JURIE F. CMML：A new metric learning approach forcross modal matching：Asian Conference on Computer Vision [EB/OL].（2019-02-07）https://www.hal.inserm.fr/GREYC/hal- 00806082v1.

[23] HU J L，LU J W，TAN Y P. Discriminative Deep Metric Learning for Face Verification in the Wild [C]//Proceedings of the2014 IEEE Conference on Computer Vision and Pattern Recognition.Washington：IEEE Computer Society，2014：1875-1882.

[24] CAI X Y，WANG C H，XIAO B H，et al. Deep nonlinear metric learning with independent subspace analysis for face verification [C]//Proceedings of the 20th ACM international conference on Multimedia.New York：Association for Computing Machinery，2012：749-752.

[25] YI D，LEI Z，LIAO S C，et al. Deep metric learning for person re-identification [C]//Proceedings of the 2014 22nd International Conference on Pattern Recognition.Washington：IEEE ComputerSociety，2014：34-39.

作者简介：

蓝凌（1978—），男，畲族，广东南雄人，高中信息技术高级教师，本科，研究方向：人工智能、机器人教育；

吴剑滨（1979—），男，汉族，广东英德人，高中信息技术高级教师（中级），本科，研究方向：高考、中考考务管理，信息化教学装备，信息化教学应用；

侯亮（1977—），男，汉族，广东韶关人，工程师，本科，研究方向：信息技术应用、视频安防。