文章基本信息

标题：Towards facial recognition using likelihood ratio approach to facial landmark indices from images
本地全文：下载
作者：Rajesh Verma ; Navdha Bhardwaj ; Arnav Bhavsar 等
期刊名称：Forensic Science International: Reports
印刷版ISSN：2665-9107
出版年度：2022
卷号：5
页码：100254
语种：English
出版社：Elsevier BV
摘要：We propose a novel approach aimed at facial comparison in forensic context. We employed an automatic approach to detect facial landmarks, and then selected independent facial indices extracted from a subset of these landmarks. As it is difficult to compare the morphometric indices due to pose variations, some statistically reliable method is required for face comparison. The present work demonstrates the use of likelihood ratios (LRs) to assess the grouping of facial images based on the morphometric indices, in a database of 40 persons, with 10 facial images each; in different poses, expressions, illumination and background. Using the criterion LR> 1, for correct grouping, and only considering indices that are not correlated, a true positive (TP) rate of 85% and false positive (FP) rate of 25% was obtained. We have also calculated the performance metrics for assessment of the validity and reliability of the likelihood ratio approach and the log likelihood ratio cost has been found to be 0.26. As compared to more abstract face identification methods, the proposed approach is relevant in forensic context, as it is based on interpretable morphometric features of faces. The likelihood ratio approach being a statistical approach is reliable as it leverages the variations due to pose variations in the face images, offers a good quality segregation of faces and hence is a promising technique for use in forensic facial identification.