A Novel Approach to Fingerprint Authentication using Histogram Oriented Gradients for Feature Extraction and Machine Learning Convolution Neural Network for Classification
Pradeep Kumar Reddy R, T N Manohara, T V Sreenivasa · International Journal of Wireless Security and Networks · 2023
With applied biometrics, it is possible to identify a person by examining a feature vector of attributes derived from their physical and behavior characteristics.In biometrics, fingerprints have become one of the most famous and well known techniques of identification and authentication.In light of technological advancements and safety, fingerprint recognition has been successfully used in a variety of Civil, Defence, and Commercial applications for more than a decade.The use of fingerprints for signature applications started in the second millennium and studies have been conducted on the different aspects as well as attributes of fingerprints over the past decade.In addition to CNN for deep learning, this study introduces a Histogram Oriented Gradient for feature extraction.Histogram equalisation, filter enhancement, and fingerprint thinning are a few of the preprocessing strategies for obtaining fingerprint features.A deep convolutional neural network algorithm has been created for the purpose of classifying preprocessed fingerprints.Compared to deep learning networks, HOG-based CNNs is extremely efficient.In addition, CNN is challenged by its inability to rotate.Despite the fact that CNN has already been investigated, We propose a novel and simple HOG-based CNN that generates highly valued data efficacy and is rotation-invariant.For validation within the database, 64 epochs achieved 99.47% accuracy, against 100% training accuracy.A validation accuracy of 4.33% was achieved for an outside database with a training accuracy of 6.33%.In comparison with contemporary machine learning algorithms, the accuracy achieved in this work is considerably higher.