摘要
目的探索一种适用于刑事案件现场勘查的曲面客体上变形指纹的校正方法,以提高疑难指纹在案件中的作用率。方法提出一种由粗到细的校正方法,以传统模式识别复原曲面畸变的结果为基础,在保证其外框架形状不变的条件下对内部图像依照神经网络校正结果进行优化,结合了传统模式识别的鲁棒性和深度卷积神经网络(deep convolutional neural network,DCNN)的精准性。结果制作了40枚模拟现场潜在变形指纹,在库容量为2000万的上海市公安局物证鉴定中心指纹测试库检索比较的实验中,所提的新型混合式校正方法效果显著优于DCNN校正方法,50名内比中率从85%上升至100%。结论该新型混合式校正方法对现场变形指纹的校正有积极意义,尤其是对低质量的变形指纹效果显著,校正后排位提升明显,有助于提高现场勘查中疑难物证的作用率。
Objective This paper put forward a method for rectifying distorted latent fingerprints left on curved objects at crime scene to improve the application rate of evidence.Methods We proposed a coarse to fine approach,modifying inner image based on fixed frame computed by traditional pattern recognition.It combines the advantages of traditional pattern recognition and deep learning network considering robustness and accuracy.Results We conducted several experiments with forty simulation samples to compare our approach with deep convolutional neural network(DCNN)based approach.The experiments usied the test database of Shanghai Public Security Bureau,which contains twenty million different fingerprints.The proposed approach presented remarkable improvement in fingerprints matching.The hit rate in top fifty candidates rose from 85%to 100%.Conclusion The proposed method is promising in handling distorted latent fingerprints,especially for low quality latent fingerprints.The ranks of genuine match rose obviously after rectification,which would improve the usage rate of distorted fingerprint evidence in crime scene investigation.
作者
高畅
沙良潇
赵雪珺
包清
GAO Chang;SHA Liangxiao;ZHAO Xuejun;BAO Qing(Criminal Investigation Department of Shanghai Public Security Bureau,Shanghai 200083,China;Shanghai Key Laboratory of Scene Evidence,Shanghai Resenrch Institute of Criminal Science and Technology,Shanghai 200083,China;University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《中国司法鉴定》
2023年第4期57-65,共9页
Chinese Journal of Forensic Sciences
基金
上海市刑事科学技术研究院现场物证重点实验室开放课题项目(2023XCWZK02)。
关键词
现场潜在指纹
变形指纹校正
混合式方法
深度卷积神经网络
latent fingerprint at scene
distortion rectification
combined method
deep convolutional neural network(DCNN)
