VGG-16 Accuracy Optimization for Fingerprint Pattern Imager Classification

Abstract

A fingerprint is a unique biometric identity commonly used as evidence in court. However, the quality of fingerprints can deteriorate due to external factors such as uneven surfaces, weather conditions, or distortion. This study uses the FVC2000 dataset and applies Convolutional Neural Networks (CNNs) to enhance and classify fingerprint images, focusing on patterns such as arches, loops, radial loops, ulnar loops, and twin loops. A novel aspect of this research is the optimization of the VGG-16 model by making specific adjustments to the hyperparameters, including setting the learning rate to 0.0001, using 50 epochs, and selecting a training-to- validation data split of 80%:10%. These adjustments were made to enhance the model’s ability to classify complex and varied fingerprint patterns, which typically present challenges to standard CNN models. The results of the study show the highest accuracy of 100% on the test data with the optimized parameters.These findings demonstrate that the optimized VGG-16 model successfully classifies fingerprint images with optimal performance. The real-world implications of achieving 100% accuracy include an increase in the reliability of biometric identification systems, especially for forensic and security applications that require high accuracy to ensure accurate decisions. This study makes a significant contribution to the development of CNN-based fingerprint classification systems, offering a new approach that supports more reliable and precise biometric applications.

Keywords

• fingerprint
• optimization
• classification
• vgg-16. cnn