Biometric score calibration for forensic face recognition

Ali, Tauseef

doi:10.3990/1.9789036536899

Cited by 8 publications

(20 citation statements)

References 70 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this paper, we have chosen three methods commonly used in forensic literature (9,42) to convert biometric scores into an LR. Methods used are the Weibull model approach (22), a parametric method that approximates two probability distribution functions (PDFs), kernel density estimation (KDE) (23), a parametric method that also generates two PDFs, and the nonparametric isotonic regression that computes a cumulative distribution function (CDF) (24).…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Likelihood Ratios for Deep Neural Networks in Face Comparison

Rodríguez

Geradts

Worring

2020

Journal of Forensic Sciences

View full text Add to dashboard Cite

In this study, we aim to compare the performance of systems and forensic facial comparison experts in terms of likelihood ratio computation to assess the potential of the machine to support the human expert in the courtroom. In forensics, transparency in the methods is essential. Consequently, state‐of‐the‐art free software was preferred over commercial software. Three different open‐source automated systems chosen for their availability and clarity were as follows: OpenFace, SeetaFace, and FaceNet; all three based on convolutional neural networks that return a distance (OpenFace, FaceNet) or similarity (SeetaFace). The returned distance or similarity is converted to a likelihood ratio using three different distribution fits: parametric fit Weibull distribution, nonparametric fit kernel density estimation, and isotonic regression with pool adjacent violators algorithm. The results show that with low‐quality frontal images, automated systems have better performance to detect nonmatches than investigators: 100% of precision and specificity in confusion matrix against 89% and 86% obtained by investigators, but with good quality images forensic experts have better results. The rank correlation between investigators and software is around 80%. We conclude that the software can assist in reporting officers as it can do faster and more reliable comparisons with full‐frontal images, which can help the forensic expert in casework.

show abstract

Section: Methodsmentioning

confidence: 99%

“…Would it be possible to obtain a valid LR in 1:1 face comparison suitable for forensics? For that end, we make use of the proceedings to attain an LR based on a biometric score (9,10). For face comparison, the biometric score is the value obtained from an automated system that can compute either the distance or dissimilarity between two given faces.…”

mentioning

confidence: 99%

Likelihood Ratios for Deep Neural Networks in Face Comparison

Rodríguez

Geradts

Worring

2020

Journal of Forensic Sciences

View full text Add to dashboard Cite

show abstract

“…This score itself has no forensic relevance and needs to be converted to LR. Four score-to-LR conversion models have been proposed [17]: Kernel Density Estimation (KDE), Linear Logistic Regression (LLR), Histogram Binning and Pool Adjacent Violators (PAV), where KDE is a commonly used method which is easy to explain. In KDE, a kernel distribution is a non-parametric representation of the probability density function (PDF) of a random variable.…”

Section: Score-to-lr Conversion Modelmentioning

confidence: 99%

Forensic face recognition based on KDE and evidence theory

Wen

2021

MATEC Web Conf.

View full text Add to dashboard Cite

Forensic face recognition (FFR) has been studied in recent years in forensic science. Given an automatic face recognition system, output scores of the system are used to describe the similarity of face image pairs, but not suitable for forensics. In this study, a score-mapping model based on kernel density estimation (KDE) and evidence theory is proposed. First, KDE was used to generate probability density function (PDF) for each dimensional feature vector of face image pairs. Then, the PDFs could be utilized to determine separately the basic probability assignment (BPA) of supporting the prosecution hypothesis and the defence hypothesis. Finally, the BPAs of each feature were combined by Dempster’s rule to get the final BPA, which reflects the strength of evidence support. The experimental results demonstrate that compared with the classic KDE-based likelihood ratio method, the proposed method has a better performance in terms of accuracy, sensitivity and specificity.

show abstract

“…Note that the prior p(H s ) in (5) is the fraction of same source pairs in the training set and it is not the prior p(H s ) set by a court of law. This process is an example of score calibration [5].…”

Section: Strength Of Evidencementioning

confidence: 99%

“…Also the eyebrow modality itself has been the topic of several studies [34,18]. In the latter study, it was shown that the eyebrow region accounts for 1 6 of the facial region while it retains 5 6 of the performance of the facial region. The remaining modalities (nose and mouth) have almost never been studied.…”

Section: Related Workmentioning

confidence: 99%

Mind the Gap: A Practical Framework for Classifiers in a Forensic Context

Zeinstra

Meuwly

Veldhuis

et al. 2018

2018 IEEE 9th International Conference on Biometrics Theory, Applications and Systems (BTAS)

View full text Add to dashboard Cite

In this paper, we present a practical framework that addresses six, mostly forensic, aspects that can be considered during the design and evaluation of biometric classifiers for the purpose of forensic evidence evaluation. Forensic evidence evaluation is a central activity in forensic case work; it includes the assessment of strength of evidence of trace and reference specimens and its outcome may be used in a court of law. The addressed aspects consider the modality and features, the biometric score and its forensic use, and choice and evaluation of several performance characteristics and metrics. The aim of the framework is to make the design and evaluation choices more transparent. We also present two applications of the framework pertaining to forensic face recognition. Using the framework, we can demonstrate large and explainable variations in discriminating power between subjects.

show abstract

Biometric score calibration for forensic face recognition

Cited by 8 publications

References 70 publications

Likelihood Ratios for Deep Neural Networks in Face Comparison

Likelihood Ratios for Deep Neural Networks in Face Comparison

Forensic face recognition based on KDE and evidence theory

Mind the Gap: A Practical Framework for Classifiers in a Forensic Context

Contact Info

Product

Resources

About