Fingerprint Identification by Training a LSTM Network with Fingerprint Segments as Sequence Inputs
B. Nithya, P. Sripriya · 2021
With the help of biometric traits, an individual can be recognized exclusively. Biometric is a human characteristic which is in-built with every individual. There are some biometric modalities which are in use like fingerprint, face, iris etc., in that, most used biometric trait is fingerprint. The biometric identification systems have challenges like intra and inter class variation, accuracy rate, false negative, false acceptance, etc. To tackle these types of challenges, instead of getting help from traditional methods, the fingerprint identification uses Neural Network concepts. This learns the internal structures of fingerprint by getting trained the fingerprint images. The main goal of this research work is to provide highest accuracy rate and to reduce false acceptance. So, the proposed model uses sequence input technique of Recurrent Neural Network (RNN) to train the images. The motivation of the work is to train the raw images and to get the highest accuracy without performing any alignment or pre-processing techniques. This work has taken the finger images from public databases of FVC2000, 2002 and 2004 to compare the proposed technique with pre-trained CNN (Convolutional Neural Network) methods. Finally, this work gives good accuracy rate than other image classification models.