A Fusion Methodology of AKAZE and Neural Network for Fingerprint Recognition

In recent years, biometric information has become essential to the maintenance of data confidentiality. In particular, fingerprints have become the most reliable biometrics system for individual human identification based on finger image characteristics. A good quality fingerprint should have at least 25 to 90 minutiae. An unclear image will result in poor recognition. In this work, we propose a novel methodology to improve fingerprint recognition. We represent the fingerprint feature using the AKAZE features. The KAZE features use nonlinear diffusion to perform local blurring on the image data while preserving the object boundaries and removing the noise. The heuristic method for calculating the matching rate is replaced with a neural network that distinguishes various data types with fewer rules. Experiments have been performed to validate the proposed method using an instance of the FVC2002 database. The proposed method has achieved adequate results for the biometric evaluation. The values of EER are less than 1.5%, with the highest success rate recorded in DB2 having an EER of 1.01%.