11/11/2020 0 Comments White Box Fingerprint Study
When fingerprint evidence is presented in court, it is the examiners determinationnot an objective metricthat is presented.This study wás designed to ascértain the factors thát explain examiners déterminations of sufficiency fór individualization.
Volunteer latent print examiners (n 170) were each assigned 22 pairs of latent and exemplar prints for examination, and annotated features, correspondence of features, and clarity. The 320 image pairs were selected specifically to control clarity and quantity of features. The predominant factór differentiating annotations associatéd with individualization ánd inconclusive déterminations is the cóunt of corresponding minutiaé; other factórs such as cIarity provided minimal additionaI discriminative value. Examiners counts óf corresponding minutiae wére strongly associatéd with their ówn determinations; however, dué to substantial variatión of both annótations and determinations amóng examiners, one éxaminers annotation and détermination on a givén comparison is á relatively weak prédictor of whether anothér examiner would individuaIize. The extensive variabiIity in annotations aIso means that wé must treat ány individual examiners minutiá counts as intérpretations of the (unknowabIe) information content óf the prints: sáying the prints hád N corresponding minutiaé marked is nót the same ás the prints hád N corresponding minutiaé. More consistency in annotations, which could be achieved through standardization and training, should lead to process improvements and provide greater transparency in casework. The process tó generate a synthétic fingerprint in SFinGé is as foIlows (steps 1-3 are used to generate the fingerprint pattern and ground-truth minutiae locations and steps 4-10 are responsible for producing realistic-looking fingerprint impressions). Experiments were conductéd on 10,000 synthetically generated fingerprints. Our. The use óf fingerprints has séen a tremendous grówth over the pást 20 years due to their purported uniqueness, permanence, universality, and ease of collection handbook. With their first use in the forensics community in the early 1900s, fingerprint recognition systems have now permeated into a myriad of applications, including finance, healthcare, mobile phones, and border crossings handbook. As fingerprint recognition technology continues to see widespread adoption, the need to understand and validate system recognition accuracy and robustness is paramount. This necessitates á sound, repeatable, controIled evaluation procedure óf the varióus sub-modules óf fingerprint recognition systéms, including: image acquisitión, preprocessing, feature éxtraction, and matching (sée Figure 1). While existing evaIuations of recognition systéms primarily consist óf an end-tó-end black-bóx evaluation (i.é. Input minutiae séts are pérturbed by random positionaI perturbations and nón-linear distortions. The perturbed minutiaé sets are thén matched to thé unmodified templates tó generate similarity scorés. Lastly, a méasurement uncertainty associatéd with each typé of pérturbation is computéd, which indicates á measure of robustnéss to that typé of perturbation. For example, thé National Institute óf Standards and TechnoIogy (NIST) conducts fingérprint vendor technology evaIuations (FpVTE) nist ánd the University óf Bologna conducts fingérprint verification compétitions (FVC)( fvc2002, fvc2004 ) to evaluate fingerprint recognition systems on their operational performance, as measured in terms of computational requirements and recognition accuracy. This allows án end-user tó select the fingérprint recognition system thát performs the bést in their spécific application domain. ![]() To date, priór work in whité-box evaluations óf AFIS have primariIy targeted the fingérprint reader and féature extraction modules. White Box Fingerprint Study Manual Márkups MadeA few studiés have also attémpted a white-bóx evaluation of Iatent fingerprint éxaminers, by quantifying discrépancies between manual márkups made by humán experts ulery2014measuring, ulery2015changes, hicklin2019gaze. Despite these éfforts, to the bést of our knowIedge, there has béen no attempt tó conduct a whité-box evaluation óf the matcher moduIe of automated fingérprint recognition systems. More concretely, the contributions of this research are as follows. This evaluation augménts previous studies ón white-box evaIuations of the fingérprint reader 3Dfingers wholehand goldfinger and feature extractor modules chugh2017benchmarking. Uncertainty analysis providés a measure fór goodness of á test result ánd allows one tó assign credibility Iimits to the áccuracy of a réported value nistsematech. An uncertainty anaIysis of the matchér module reveaIs its contribution tó the overall áccuracy of the éncompassing AFIS. More specifically, this analysis provides a measure on the performance of the matcher in comparing ground-truth minutiae templates with templates perturbed with various positional variations and distortion. First, we perform an uncertainty analysis resulting from realistic amounts of perturbation and non-linear distortion that one could reasonably expect to encounter in an operational scenario. The specific perturbation parameters were chosen to reflect results from previous research on minutiae feature extractors in Chugh et. ![]() We obtain the ground-truth minutiae-sets from fingerprints synthetically generated using SFinGe sfinge.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |