da/sec scientific talk on Biometrics
Topic: Fingerprint Image Quality – Predicting Biometric Performance
Keywords — fingerprint recognition, image quality, biometric performance, image processing, machine learning
Biometric systems and fingerprint recognition systems in particular have become very widespread in recent years, both in mobile devices and through increased usage in border controls and electronic national IDs. A crucial aspect in any biometric system is that the quality of the data that enters system is of the highest possible quality to facilitate ease of interaction, robust system performance and a high level of biometric performance.
The predictive performance of state of the art quality assessment algorithms is evaluated and illustrate that the predictive capabilities vary widely across dataset and chosen comparison subsystem. A dataset collection performed under controlled environmental conditions expose differences between current and previous generation of fingerprint sensors with respect to the resistance towards skin moisture. Application of algebraic topology gives new perspectives on the fundamental structure of fingerprint images and enable distinguishing them from non-fingerprint images with lower error than current state of the art. Working towards a system which assists dactyloscopic examiners in the assessment of fingerprint quality and determination of evidential value for forensic uses it is shown that examiners assigned quality scores are indicative of comparison scores. Towards quality assessment on devices with limited computational resources, a method based on a combination of clustering using receptive fields and random forests is proposed and found to result in predictive performance comparable to state of the art methods.