Fingerprint Recognition Using a Transfer Learning Method

Abstract

Individuals' recognition has been the major concern of many computer science scholars. This paper highlights the development of a biometric fingerprint recognition system using an artificial intelligence method. The methodology used is structured into three main steps: fingerprint image acquisition, feature extraction and comparison or matching. Firstly, we use non-contact fingerprint images for training the recognition model to allow the developed system to work without contact. Secondly, we used a database of nineteen individuals each having fifteen fingerprint images acquired without contact. To perform the transfer learning, we have exploited the MobileNets model while adapting its architecture to the new task of contactless fingerprint classification. After training the new model obtained, we performed it evaluation through a confusion matrix. This evaluation reveals that the developed method confuses one individual out of 19, i.e. a confusion rate of 5.26%. This rate testifies to the efficiency of the method used for non-contact fingerprint recognition.