Description of Invention
This study introduces a novel few-shot learning (FSL) method for facial skin type analysis, enabling accurate classification with limited labeled data. A diverse dataset of facial images representing various skin tones and conditions was assembled and annotated. Leveraging pre-trained deep neural networks and FSL algorithms like prototypical networks (PNs) and matching networks (MNs), this approach addresses data scarcity. The research has significant implications for enhancing access to dermatological care, especially among underserved populations. It allows individuals to identify their skin type swiftly and tailor skincare routines accordingly. Experimental results indicate PNs achieving 95.78% accuracy in a 2-way, 10-shot, 15-query scenario, while MNs reached 90.33% accuracy in a 2-way, 5-shot, 10-query scenario. In conclusion, this study underscores FSL potential to improve facial skin analysis beyond traditional methods, opening avenues for further research.