Computation efficiency for core-based fingerprint recognition algorithm

This paper presents a new computationally efficient fingerprint algorithm for automatic recognition (CEFAR). The algorithm uses 41% of the original fingerprint image for recognition and reduces more than 60% of the computations for detecting the singularity point (SP). Analytical results have shown that the CEFAR maintains high accuracy over benchmark algorithms, EER less than 5% and average accuracy improvement is 337.7%, together with dramatic reduction in computation steps, leading to an efficient performance. CEFAR uses the Gabor filter to enhance the fingerprint image after performing pre-processing operations, such as segmentation and normalisation. Fingerprint features are then extracted with reference to the SP forming what is called the star structure using the Conditional Number concept that is applied to the skeleton of the fingerprint. This structure is invariant with respect to global rotations and translations on the fingerprint due to the consistency of its formation, which benefits the fingerprint matching procedure. For SP detection, core type was detected by using complex filtering applied to the orientation tensor field; this algorithm has been modified to reduce computational complexity, although it has a high accuracy performance, where results have shown that more than 95% of the SP's have been successfully detected.