Regularized Pyramid Convolutional Layers for Human Skin Type Assessment

Skin type classification is significant in both dermatology and computer vision, serving in personalized skincare recommendations, disease diagnosis, and maintaining algorithmic fairness. The complex non-linear relationships among various skin characteristics, such as texture, color, elasticity, and sensitivity, poses challenges for traditional techniques and machine learning models in capturing the features effectively. In response to this challenge, this research introduces an innovative approach to enhance human skin type analysis through the application of Pyramid Convolutional Layers (PCL). Differing from conventional methods, this study addresses the limitations of existing models in accurately capturing the intricate relationships between different skin attributes. The Fitzpatrick17k dataset, comprising 16,577 clinical images from DermaAmin and Atlas Dermatologico, annotated using Fitzpatrick Scale labels through Scale AI and Centaur Labs, forms the foundational basis of the investigation. According to implications in dermatology, Pyramid Convolutional Layers were specifically chosen to regularize and capture multi-scale and contextual attributes, thereby enhancing classification accuracy. In the preprocessing stage, images undergo standardization through uniform resizing and center-cropping, aligning with a classification-centric approach. It found that the efficacy of the proposed method revealing significant accuracy improvements in skin type classification compared to traditional method. Comprehensive comparative analyses establish the superior performance of our architecture, emphasizing its potential to advance human skin type classification accuracy. This study contributes to the field of dermatological image analysis and highlights the broader applications of Pyramid Convolutional Layers, showcasing methodological advances at the intersection of computer vision and dermatology.