A lightweight CNN model for binary classification of fingerprint images

Abstract

Fingerprint recognition is one of the most widely used biometric identification techniques due to its uniqueness and long-term stability. However, intentional fingerprint alterations can significantly reduce the reliability of automatic recognition systems, creating a need for robust classification methods capable of distinguishing between genuine and altered fingerprint images. This paper presents a lightweight Convolutional Neural Network model for binary fingerprint classification. The proposed model consists of four convolutional layers with progressively increasing numbers of filters, combined with batch normalization, ReLU activation, max-pooling, global average pooling, and fully connected layers, providing a balance between classification performance and computational complexity. The study also examines the influence of class weighting and input image resolution on model performance. Experimental results demonstrate that the proposed model achieved a classification accuracy of 94.3%, with high precision, recall, and F1-score values. In addition, the results indicate that the class weighting technique significantly improves the classification of minority-class samples, whereas increasing the input image resolution does not substantially affect the overall classification performance. These findings suggest that the proposed lightweight CNN architecture can provide reliable performance for binary fingerprint classification while maintaining relatively low computational requirements. This indicates the potential of the proposed model for biometric applications and possible applicability within industrial systems, particularly in secure access control and authentication scenarios where computational efficiency is an important consideration.