Advancing Skin Type Classification: The Development of Malaysian Facial Skin Texture Datasets for Machine Learning

Accurate facial skin type classification is crucial in dermatology and cosmetics, with systems like the Baumann Skin Type System (BSTS) offering a comprehensive framework. However, no publicly available dataset supports facial skin type classification for research purposes. Existing public datasets, such as those for skin diseases, focus on medical conditions and lack the granularity needed for general skin type studies. Attempts to collaborate with dermatological institutions to access proprietary datasets were unsuccessful due to Health Insurance Portability and Accountability Act (HIPAA) regulations and consent issues, further highlighting the resource gap. This paper presents the development of the Malaysian Facial Skin Texture Dataset (MFSTD) in two iterations, focusing on leveraging Few-Shot Learning (FSL) to optimize Baumann Skin Type classification. MFSTD version one (MFSTDv1) provided a balanced dataset under controlled and uncontrolled environments, while MFSTD version two (MFSTDv2) introduced improved diversity, annotations, and preprocessing methods. Initial experiments demonstrated that both MFSTDv1 and MFSTDv2 achieve high classification accuracy, with MFSTDv1 reaching <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$94.23 \pm 3.23 \%$</tex> in the 3-way, 10-shot, 5-query scenario and MFSTDv2 achieving <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$92.51 \pm 0.28 {\%}$</tex> in the 3 -way, 10 -shot, 10 -query scenario, showcasing their suitability for machine learning applications, particularly for FSL techniques. These results underline the potential of MFSTD as a valuable resource for advancing facial skin type classification.