Enhancing Biometric Identification

This chapter proposes a novel way to address the issues of toddler iris recognition in biometric identification. In connection with gathering information, the CASIA Iris V3 dataset is utilized to acquire images. At the same time, preprocessing comprises the use of Generative Adversarial Networks (GANs) for denoising and iris image augmentation to improve input quality. Dense U-Net integration makes it quicker to segment the iris and does it efficiently, resulting in a Dice ratio of 0.96. Swin Transformer is used in feature extraction to extract features with a global sensitivity of 95.2%. Classification involves using local patterns and triplets, which results in a loss of function precision where the pattern matches 96.0% of the time while improving the F1 score and minimizing adversarial hazards. The decision-making step is therefore shown to have an accuracy of 99.6%, an Equal Error Rate (EER) of 0.6%, and an Area Under the Curve (AUC) of 97.6%. These findings thus position an AI-based biometric system as a critical component in high-risk domains such as child protection and anti-trafficking, while also ensuring that the technology is used ethically and securely for vulnerable populations.