Water loss and hydration estimation for skin barrier function evaluation

Abstract

Everyone desires hydrated, lustrous skin. Skin with high hydration not only appears younger, but also provides a better canvas for cosmetics. However, how can one accurately determine the hydration status of their skin? To date, the popular method on the market is based on the principle of capacitance, but this method requires specific instruments and is not portable. A more convenient monitoring approach is to use ordinary selfies analyzed through machine learning to assess hydration levels and trans-epidermal water loss. Due to the limitations of datasets, standard machine learning models cannot meet the analysis needs of selfies under various lighting conditions. This dissertation proposes a novel method that employs domain generalization techniques to enhance the predictive performance of models trained with limited selfie data. Specifically, we first utilize the Retinex preprocessing technique and then improve the model through comparative learning to improve the accuracy of the model. Finally, we fully evaluate the training results and conclude this dissertation with a discussion of future research directions and improvement methods.