JPEG optimisation for fingerprint recognition: generalisation potential of an evolutionary approach

Abstract

For fingerprint-based biometric systems, JPEG has traditionally been one of the primary storage formats. In this paper, we investigate methods of optimising JPEG compression for increased matching scores at constant compression rates. To achieve this, an evolutionary approach is employed to optimise the quantisation matrix used by JPEG. While finding matrices that are better suited for different sets of fingerprints, different fingerprint sensors, and different recognition algorithms as they have been optimised for remains elusive, we find improvements in matching performance for dataset fitting scenarios.