Reliable Fingerprint Classification Based on Novel Deep Learning Approach
Shashi Kant Gupta, Ahmed Al‐Emran, Christodoss Prasanna Ranjith, M. Syed Khaja Mohideen
Abstract
In particular during the detection phase, biometric categorization is crucial to fingerprint classification (FC). In reality, when working with massive datasets, reducing the occurrence of similarities in fingerprint verification systems is vital. By dividing fingerprints into many types, the FC seeks to accomplish this goal. Image processing is only one of the numerous fields where deep learning (DL) is being used to great effect. In this study, we provide attention weighted inception-resnet-v2 (A-IRV2) approach to reliable FC. Since there are noises and redundant samples, we first analyze the PolyU dataset for this study, and this must be pre-processed. The sparse autoencoder (SA) is used to perform the pre-processing. We draw the conclusion from the experiments conducted that the suggested method performs more efficiently than other methods in terms of accuracy, precision, sensitivity, and f1-score criteria. The fingerprint images were therefore divided into clear, unclear, and damaged image outputs via A-IRV2.