Abstract
New approaches are introduced for the fundamental components of a fingerprint reconstruction system, namely, fingerprint segmentation, scratch detection and, orientation smoothing in large degraded regions. The orientation information in corrupted regions is restored without relying on the singular points. The simulation results show that the error between the predicted orientation fields and the ground truth values is negligible in terms of Mean Absolute Error, and is comparable with the error produced by other competitive reconstruction methods. The computer simulations are conducted on 180 images of FVC2004 DB3_A and 80 images FVC2000 DB1_B databases of fingerprint images. The presented tools are tested using a Fingerprint Recognition System and demonstrated a great ability in improving the Receiver Operating Characteristic curve and Equal Error Rate.