Implementation Contactless Palmprint Recognition System

Palmprint recognition is a promising biometric approach for personal identification due to its unique and stable features. This study presents a deep learning-based palmprint recognition system utilizing the ResNet50 convolutional neural network. A comparative analysis was conducted between two training strategies: training the model from scratch and performing fine-tuning on a pretrained model via transfer learning. The models were evaluated on a curated dataset comprising +11.200 contactless palmprint images from 30 distinct user classes. The fine-tuned ResNet50 model significantly outperformed the scratch-trained model, achieving an accuracy of 99.65% compared to 98.59%. The fine-tuned approach also demonstrated superior generalization and stability, yielding macro-averaged precision, recall, and F1-scores of 0.9979, 0.9987, and 0.9983, respectively. These results indicate a more robust performance with fewer misclassifications, particularly in challenging or less-represented classes. The research culminates in the successful deployment of the system through a real-time web-based interface, providing a complete end-to-end implementation for practical application.