A Cascaded Fingerprint Quality Assessment Scheme for Improved System Accuracy
Poor-quality images mostly result in spurious or missing features, which further degrade the overall performance of fingerprint recognition systems. This paper proposes a reconfigurable scheme of quality checks at two different levels: i) at raw image level and ii) at feature level. At first level, ellipse properties are calculated through analysis of statistical attributes of the captured raw image. At second level, the singularity points (core & delta) are identified and extracted (if any). These information, as quality measures, are used in a cascaded manner to block/pass the image. This model is tested on both publicly available (Cross Match Verifier 300 sensor) as well as proprietary (Lumidigm Venus V100 OEM Module sensor) fingerprint databases scanned at 500 dpi. The experimental results show that this cascaded arrangement of quality barricades could correctly block poor quality images and hence elevated the overall system accuracy: with quality checks, both FNMR and FMR significantly dropped to 9.52% & 0.26% respectively for Cross Match Dataset and 2.17% & 2.16% respectively for Lumidigm Dataset.
Keywords: Fingerprint Image Quality, Elliptical ROI, Singularity Points, Euclidean Distance
Download Full-Text
