Deep learning based forensic face verification in videos
Jinhua Zeng, Jinfeng Zeng, Xiulian Qiu · 2017
Deep learning for face identification-verification application has been proven to be fruitful. Human faces constituted the main information for human identification besides gait, body silhouette, etc. Deep learning for forensic face identification could provide quantitative indexes for face similarity measurement between the questioned and the known human faces in cases, which had the advantage of result objectivity without expert experience influences. We studied the deep learning based face representation for forensic verification of human images. Its application strategies and technical limitations were discussed. We proposed a “winner-take-all” strategy in the case of the forensic identification of human images in videos. We expected the theories and techniques for forensic identification of human images in which both qualitative and quantitative analysis methods were included and expert judgment and automatic identification methods were coexisted.