Abstract:
The proliferation of biometric systems has highlighted the importance of fingerprint classification, a critical module in fingerprint recognition systems. The rapid growth in the volume of biometric data requires solutions that can balance efficiency and accuracy in classification particularly in resource-constrained environments. Although deep learning models are efficient for fingerprint classification, their high computational demands, render them unsuitable for use in resource-constrained environments. Furthermore, the complexity of poor-quality fingerprint images poses significant challenges to achieving high accuracy. This research proposes a novel lightweight learning model tailored for fingerprint classification in content-based image retrieval systems. The combination of lightweight models and attention modules in classification tasks has not been explored extensively. By truncating a selected deep model—removing deeper, computationally intensive layers—we reduce complexity while maintaining performance accuracy. Truncation points were determined experimentally, starting with deeper layers that are more computationally extensive and have abstract features. Each truncation experiment involved removing a number of layers, performing classification tasks, and recording the results. To lessen the potential performance loss from truncation, we integrate attention modules to selectively focus on important areas of the fingerprint image. Experimental evaluation using benchmark datasets shows that the proposed model achieves good and acceptable results: 96.3% accuracy, 96.3% precision, 96.3% recall, and a 96.2% F1 score, all while significantly reducing computational overhead and training parameters. In many biometric applications, classification accuracies above 95% are often viewed as strong, especially for systems that deal with fingerprints data. The integration of attention modules within a lightweight learning model offers a promising solution for high-accuracy, low-cost fingerprint classification, particularly in resource-constrained environments. This research contributes to the field of deep learning models by presenting an innovative lightweight model for fingerprint classification that balances efficiency and accuracy.