Sclera-based biometric for age group estimation using a hybrid ResNet-50 and gradient boosted neural network

Abstract

Age estimation using biometric features, particularly sclera biometrics, has emerged as a promising alternative to facial recognition in scenarios with obscured facial features, garnering attention in the fields of computer vision and security. This study proposed a hybrid sclera based biometric model combining ResNet - 50 for feature extraction and a Gradient Boosting Neural Network (GBNN) for classification to improve age group estimation. A dataset of 2,400 sclera images from 300 subjects, categorized into four age groups ( Children: 0 – 12, Teens: 13 – 19, Young Adults: 20 – 39, Older Adults: 40+ ) was utilized, sourced from Nigerian populations to ensure regional representation and preprocessed using Otsu thresholding for precise sclera segmentation. Evaluated on a 60/20/20 train validation test split, the model achieved 94.15% accuracy, outperforming standalone ResNet - 50 (92.08%). The GBNN’s iterative refinement reduced misclassifications in challenging categories (e.g., Children) by 50%, attributed to its ability to learn residual errors and adaptively weight underrepresented samples. These results highlight the potential for deployment in biometric security systems and age restricted content moderation, particularly in low light environments where sclera visibility is consistent. While geographic homogeneity in the dataset and illumination dependent on Otsu thresholding represent current limitations, this work establishes a foundation for future advancements in multimodal biometric integration and cross population validation.