基于人脸特征相似度分数似然比的人脸比对方法
|
摘要
在法庭科学中,特征比对是进行物证检验的核心方法之一,应用于几乎所有专业。基于统计框架的特征比对客观方法,是当前法庭科学发展的方向。本文就影像专业的人脸特征比对方法展开研究。通过深入分析当前基于深度学习的人脸特征进行比对的特点,开展了大规模数据的特征比对实验,统计了深度学习特征比对分数的分布,结合贝叶斯统计框架下基于分数似然比的模型,提出基于深度学习特征相似度分数似然比的人脸比对方法。我们的实验结果和分析,支撑了人脸特征比对客观方法的实际应用,也丰富了基于统计的法庭科学特征比对方法。
Feature-comparison is one of the core methods among forensic evidence test, almost being applied by every professional subject. The feature-comparison method, based on the statistical framework, is objective, thus becoming the on-going direction of forensic science. Facial feature comparison is explored in this paper. Through in-depth characteristic analysis of the current deep learning with face features, the facial feature comparison is carried out into relevant large-scale data, thereby having obtained the statistical distributions of facial feature comparison score by deep-learning. Accordingly, the facial comparison approach is proposed at the basis of features' deep-learning coupled into the model of score-based likelihood ratio under Bayesian framework. The experimental results are supportive for the facial feature comparison to apply, demonstrating one more enrichment of the methods about forensic feature comparison based on statistics.
Feature-comparison is one of the core methods among forensic evidence test, almost being applied by every professional subject. The feature-comparison method, based on the statistical framework, is objective, thus becoming the on-going direction of forensic science. Facial feature comparison is explored in this paper. Through in-depth characteristic analysis of the current deep learning with face features, the facial feature comparison is carried out into relevant large-scale data, thereby having obtained the statistical distributions of facial feature comparison score by deep-learning. Accordingly, the facial comparison approach is proposed at the basis of features' deep-learning coupled into the model of score-based likelihood ratio under Bayesian framework. The experimental results are supportive for the facial feature comparison to apply, demonstrating one more enrichment of the methods about forensic feature comparison based on statistics.
引文
[1]PCAST.Report on forensic science in criminal courts:ensur-ing scientific validity of feature-comparison methods[R/OL].(2016-09-20)[2018-10-12].https://obamawhitehouse.archives.gov/sites/default/files/microsites/ostp/PCAST/pcast_forensic_science_report_final.pdf
[2]FSR.Forensic image comparison and interpretation evidence:guidance for prosecutors and investigators,issue 2[R].Birming-ham:Forensic Science Regulator,2016.
[3]NEUMANN C,EVETT I W,SKERRETT J E,et al.Quantify-ing the weight of evidence from a forensic fingerprint comparison:a new paradigm[J].Journal of The Royal Statistical Society Series A-statistics in Society,2012,175(2):371-415.
[4]CHAMPOD C,EVETT I W.A probabilistic approach to finger-print evidence[J].Journal of Forensic Identifica-tion,2001,51(2):101-122.
[5]HEPLER A,SAUNDERS C P,DAVIS L J,et al.Score-based likelihood ratios for handwriting evidence[J].Forensic Science International,2012,219(1):129-140.
[6]MEUWLY D,DRYGAJLO A.Forensic speaker recognition based on a Bayesian framework and Gaussian mixture model-ing[J].Forensic Science International.2003,136(Suppl.1):364.
[7]WALSH K,BUCKLETON J S,TRIGGS C M.A practi-cal example of glass interpretation[J].Science&Justice,1996,36(4):213-218.
[8]NATIONAL RESEARCH COUNCIL.Reference manual on scientific evidence[M].3rd ed.Washington,DC:The National Academies Press,2011
[9]BUCKLETON J S,TRIGGS C M,WALSH S J.Forensic DNAevidence interpretation[M].Florida:CRC Press,2005.
[10]AITKEN C,ROBERTS P,JACKSON G.Fundamentals of probability and statistical evidence in criminal proceedings,Guidance for Judges,Lawyers,Forensic Scientists and Expert Witnesses[M].Royal Statistical Society,2010.
[11]张翠玲,谭铁君.基于贝叶斯统计推理的法庭证据评价[J].刑事技术,2018,43(4):265-271.
[12]MEUWLY D,RAMOS D,HARAKSIM R,et al.A guideline for the validation of likelihood ratio methods used for foren-sic evidence evaluation[J].Forensic Science International,2017,276(7):142-153.
[13]ENFSI.ENFSI guideline for evaluative reporting in forensic science:strengthening the Evaluation of Forensic Results across Europe(STEOFRAE)[M].Wiesbaden,Ger:European Network of Forensic Science Institutes,2015
[14]GONZALEZ-RODRIGUEZ J,FIERREZ-AGUILARJ,RAMOS-CASTRO D,et al.Bayesian analysis of fingerprint,face and signature evidences with automatic biometric systems[J].Forensic Science International,2005,155(2-3):126-140.
[15]MEUWLY D,GOODE A,DRYGAJLO A,et al.Forensic speaker recognition based on a Bayesian framework and Gaussian mixture modeling[J].Forensic Science International,2003,136(Suppl.1):364.
[16]ALLEN R.Exact solutions to Bayesian and maximum likeli-hood problems in facial identification when population and error distributions are known[J].Forensic Science International,2008,179(2):211-218.
[17]ALI T,SPREEUWERS L,VELDHUIS R,et al.Effect of cali-bration data on forensic likelihood ratio from a face recogni-tion system[C]//.IEEE.IEEE Sixth International Conference.Institute of Electrical and Electronics Engineers,2013:1-8.
[18]MERY D,ZHAO Y N,BOWYER K.On the Reproducibility and Repeatability of Likelihood Ratio in Forensics:A case study using Face Biometrics[C]//IEEE.The IEEE Eighth In-ternational Conference on Biometrics:Theory,Applications,and Systems(BTAS 2016).Institute of Electrical and Electron-ics Engineers,2016.
[19]DAVIS L J,SAUNDERS C P,HEPLER A,et al.Using subsampling to estimate the strength of handwriting evidence via score-based likelihood ratios[J].Forensic Science Internation-al,2012,216(1):146-157.
[20]KLEINBERG K F,VANEZIS,VANEZIS P.BURTON A M.Failure of anthropometry as a facial identification technique using high-quality photographs[J].Journal of Forensic Sci-ences 2007,52(4):779-783.
[21]杜达,哈特,斯多克.模式分类[M].李宏东,姚天翔,译.北京:机械工业出版社,2003.
[22]SUN Y,WANG X,TANG X,et al.Deep Learning Face Rep-resentation from Predicting 10,000 Classes[C]//.IEEE.2014IEEE Conference on computer vision and pattern recognition.Institute of Electrical and Electronics Engineers,2014:1891-1898.
[23]SUN Y,WANG X,TANG X,et al.Deep Convolutional Net-work Cascade for Facial Point Detection[C]//.IEEE.2013IEEE Conference on computer vision and pattern recognition.Institute of Electrical and Electronics Engineers,2013:3476-3483.
[24]HINTON G E,SRIVASTAVA N,KRIZHEVSKY A,et al.Im-proving neural networks by preventing co-adaptation of feature detectors[J].arXiv:Neural and Evolutionary Computing,2012.
[25]中华人民共和国公安部.安防人脸识别应用系统第2部分人脸图像数据:GA/T 922.2-2011[S].北京:中国标准出版社,2011.
[26]中华人民共和国公安部.居民身份证制证用数字相片技术标准:GA461-2004[S].北京:中国标准出版社,2004.
[2]FSR.Forensic image comparison and interpretation evidence:guidance for prosecutors and investigators,issue 2[R].Birming-ham:Forensic Science Regulator,2016.
[3]NEUMANN C,EVETT I W,SKERRETT J E,et al.Quantify-ing the weight of evidence from a forensic fingerprint comparison:a new paradigm[J].Journal of The Royal Statistical Society Series A-statistics in Society,2012,175(2):371-415.
[4]CHAMPOD C,EVETT I W.A probabilistic approach to finger-print evidence[J].Journal of Forensic Identifica-tion,2001,51(2):101-122.
[5]HEPLER A,SAUNDERS C P,DAVIS L J,et al.Score-based likelihood ratios for handwriting evidence[J].Forensic Science International,2012,219(1):129-140.
[6]MEUWLY D,DRYGAJLO A.Forensic speaker recognition based on a Bayesian framework and Gaussian mixture model-ing[J].Forensic Science International.2003,136(Suppl.1):364.
[7]WALSH K,BUCKLETON J S,TRIGGS C M.A practi-cal example of glass interpretation[J].Science&Justice,1996,36(4):213-218.
[8]NATIONAL RESEARCH COUNCIL.Reference manual on scientific evidence[M].3rd ed.Washington,DC:The National Academies Press,2011
[9]BUCKLETON J S,TRIGGS C M,WALSH S J.Forensic DNAevidence interpretation[M].Florida:CRC Press,2005.
[10]AITKEN C,ROBERTS P,JACKSON G.Fundamentals of probability and statistical evidence in criminal proceedings,Guidance for Judges,Lawyers,Forensic Scientists and Expert Witnesses[M].Royal Statistical Society,2010.
[11]张翠玲,谭铁君.基于贝叶斯统计推理的法庭证据评价[J].刑事技术,2018,43(4):265-271.
[12]MEUWLY D,RAMOS D,HARAKSIM R,et al.A guideline for the validation of likelihood ratio methods used for foren-sic evidence evaluation[J].Forensic Science International,2017,276(7):142-153.
[13]ENFSI.ENFSI guideline for evaluative reporting in forensic science:strengthening the Evaluation of Forensic Results across Europe(STEOFRAE)[M].Wiesbaden,Ger:European Network of Forensic Science Institutes,2015
[14]GONZALEZ-RODRIGUEZ J,FIERREZ-AGUILARJ,RAMOS-CASTRO D,et al.Bayesian analysis of fingerprint,face and signature evidences with automatic biometric systems[J].Forensic Science International,2005,155(2-3):126-140.
[15]MEUWLY D,GOODE A,DRYGAJLO A,et al.Forensic speaker recognition based on a Bayesian framework and Gaussian mixture modeling[J].Forensic Science International,2003,136(Suppl.1):364.
[16]ALLEN R.Exact solutions to Bayesian and maximum likeli-hood problems in facial identification when population and error distributions are known[J].Forensic Science International,2008,179(2):211-218.
[17]ALI T,SPREEUWERS L,VELDHUIS R,et al.Effect of cali-bration data on forensic likelihood ratio from a face recogni-tion system[C]//.IEEE.IEEE Sixth International Conference.Institute of Electrical and Electronics Engineers,2013:1-8.
[18]MERY D,ZHAO Y N,BOWYER K.On the Reproducibility and Repeatability of Likelihood Ratio in Forensics:A case study using Face Biometrics[C]//IEEE.The IEEE Eighth In-ternational Conference on Biometrics:Theory,Applications,and Systems(BTAS 2016).Institute of Electrical and Electron-ics Engineers,2016.
[19]DAVIS L J,SAUNDERS C P,HEPLER A,et al.Using subsampling to estimate the strength of handwriting evidence via score-based likelihood ratios[J].Forensic Science Internation-al,2012,216(1):146-157.
[20]KLEINBERG K F,VANEZIS,VANEZIS P.BURTON A M.Failure of anthropometry as a facial identification technique using high-quality photographs[J].Journal of Forensic Sci-ences 2007,52(4):779-783.
[21]杜达,哈特,斯多克.模式分类[M].李宏东,姚天翔,译.北京:机械工业出版社,2003.
[22]SUN Y,WANG X,TANG X,et al.Deep Learning Face Rep-resentation from Predicting 10,000 Classes[C]//.IEEE.2014IEEE Conference on computer vision and pattern recognition.Institute of Electrical and Electronics Engineers,2014:1891-1898.
[23]SUN Y,WANG X,TANG X,et al.Deep Convolutional Net-work Cascade for Facial Point Detection[C]//.IEEE.2013IEEE Conference on computer vision and pattern recognition.Institute of Electrical and Electronics Engineers,2013:3476-3483.
[24]HINTON G E,SRIVASTAVA N,KRIZHEVSKY A,et al.Im-proving neural networks by preventing co-adaptation of feature detectors[J].arXiv:Neural and Evolutionary Computing,2012.
[25]中华人民共和国公安部.安防人脸识别应用系统第2部分人脸图像数据:GA/T 922.2-2011[S].北京:中国标准出版社,2011.
[26]中华人民共和国公安部.居民身份证制证用数字相片技术标准:GA461-2004[S].北京:中国标准出版社,2004.