An Interpretable Palmprint Recognition Approach Using a CNN and Explainable-AI Frameworks

This study implemented a biometric authentication workflow using contactless palmprint recognition. With the rise of artificial intelligence (AI) for this application, the need for transparent and interpretable models has become crucial due to their black-box nature. An end-to-end lightweight convolutional neural network (CNN) was developed, and two explainable AI frameworks, Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) were used to incorporate and enhance model interpretability. The methodology included training the model on multispectral contactless palmprint images after comprehensive pre-processing steps, such as region of interest (ROI) extraction, denoising using Non-local Means (NLMeans), and image augmentation. The model's feature learning was then validated using LIME and SHAP, which helped uncover key features influencing the model's decisions. The CNN was then evaluated on an unseen testing set, demonstrating high identification accuracy and outperforming a pre-trained model approach. This work highlights the role of XAI in developing reliable and trustworthy biometric authentication systems.