Robust skin type classification using convolutional neural networks

Skin spectrum is used in a wide range of applications including medical science, dermatology, cosmetics science, and biometric face recognition. However, it is noticed that the composition of complex tissue layers and the uneven outer surface of the skin make skin spectrum evaluation error-prone. In other words, the skin reflection spectra of the same measurement area from the same person could show different spectral characteristics. Recently, Deep Learning algorithms show robust classification results in the area such as visual recognition, image labeling, speech recognition, and hyperspectral image. In this work, a commonly used Deep Learning method, Convolutional Neural Network, is introduced for studying robust Fitzpatrick skin type classification. Considering the small sample size of the skin spectra dataset in this paper, a single convolutional layer Convolutional Neural Network model is applied. To evaluate the performance of our simplified Convolutional Neural Network model, an Artificial Neural Network model, as well as the traditional ITA Fitzpatrick classification approach are also compared. The classification result of our Convolutional Neural Network model shows a better Fitzpatrick skin type classification, with an accuracy rate up to 92.59%.