De-convolutional auto-encoder for enhancement of fingerprint samples
Patrick Schuch, Simon Schulz, Christoph Busch · 2016
Reliability and accuracy of the features extracted from fingerprints are essential for the performance of any fingerprint comparison algorithm. Image Enhancement as a pre-processing step allows to extract features more accurately by enhancing the quality of the fingerprint signal. This work proposes to use De-Convolutional Auto-Encoders for fingerprint image enhancement. Its performance is compared to seven state-of-the-art methods regarding their improvements for recognitions of the biometric system. Biometric performance is tested with MINDTCT and FingerJetFX for feature extraction and BOZORTH3 for biometric comparison. Critical comparisons are determined from 14 datasets. Those are used for evaluation of the methods. The impact of a method on biometric performance varies significantly. No single image enhancement can be found, which works best for all combinations. However, the proposed method ConvEnhance achieves highest count of best improvements among the evaluated methods.