HRSTA: A Pipeline for Reference-Based High-Resolution Skin Tone Adjustment
Boqi Wu, Ulrich Jung, Hasan Tercan · 2024
Different skin tones reflect the diversity of the world's population.In the dynamic global retail environment, especially in the fashion industry, there is a growing need to adapt marketing strategies to authentically represent different ethnicities, thereby increasing consumer engagement and the likelihood of purchase.However, creating photo shoots that include a wide range of skin tones is expensive and time-consuming.Moreover, human visual perception is particularly adept at noticing the fine details and natural appearance of skin tones.Our study presents an Artificial Intelligence (AI)-based solution for cost-effectively modifying skin tones in images.We propose a High-Resolution Skin Tone Adjustment (HRSTA) pipeline that employs segmentation models, Generative Adversarial Networks (GANs), and digital image processing techniques to adjust skin tones to match references under varying lighting conditions, without compromising image resolution or authenticity.Our experiments demonstrate that HRSTA is a robust baseline for skin tone adjustment, offering a promising solution for fashion, beauty and other industries.