Facial skin colour classification using machine learning and hyperspectral imaging data

Facial skin colour is a primary indicator of human health, and skin care products aim to improve it. However, the beauty industry lacks accurate methods to automatically classify skin colour because it uses high granularity labels such as the Pantone scale, with up to 110 facial skin-colour types. In this study, the automatic classification of facial skin colour based on hyperspectral imaging and machine learning is investigated. An experiment is conducted using hyperspectral imaging to collect multi-dimensional big data on most of these skin-colour types. Owing to the multi-dimensionality of the data and the high granularity of skin-colour types, classifying the colour type accurately is challenging. Nevertheless, various machine-learning methods are applied and it is found that each had an advantage in categorizing a subset of colour types. Further, the features of the data are analysed and it is found that skin chromaticity and brightness could be classified separately to provide valuable information. Finally, to utilise this information and combine the advantages of various machine-learning methods, a two-stage integrated classifier is developed, which achieved 90.4% accuracy. This classifier can be used by the beauty industry to evaluate the effect of skincare products on facial skin colour.