Enhanced palmprint recognition using a hybrid deep learning framework with Chebyshev layer integration

A palmprint, a small section of the palm's surface, contains information that can be used for authentication systems. It is also characterized by permanence, meaning it does not change over time. Extracting meaningful features from palmprints is essential, as traditional methods relying on primary lines, wrinkles, and creases are often insufficient to distinguish between individuals due to their proximity. Deep learning techniques are increasingly employed to extract more profound features, such as texture. A hybrid deep learning framework for enhanced palmprint recognition is proposed, which combines a pre-trained ResNet-18 model with novel layers, such as the Chebyshev layer, to improve accuracy and performance in palmprint classification tasks. The integration of the Chebyshev layer serves as a key differentiator for the model, enhancing its ability to extract more complex and robust features from palmprint images. This model was tested using both Chinese Academy of Sciences Institute of Automation (CASIA), and Touchless datasets (IIT-Delhi), achieving an accuracy of 99.98%, F1-score of 99.81%, and a precision of 100% on IIT-Delhi dataset and accuracy of 99.26%, F1-score of 99.26%, and a precision of 100% on CASIA dataset. The ROC-AUC values are 100%and 99.81%, respectively, while the EER values remain below 0.3% for both datasets. The results of the performance comparison conclude that the suggested model outperforms others by achieving the highest accuracy, which proves its effectiveness in palmprint-based identity recognition.