Skin Cancer Data Augementation Method Based on Color Transfer

There are eight common types of skin cancer. Using deep learning is a popular way of helping users to classify whether their lesion is benign or malignant. If they could observe their disease as soon as possible, it can improve patients’ survival chances. However, the datasets for the medical field have joint problems. That is, the size of the datasets is rare. The popular skin cancer dataset that we use frequently has few apparent issues. The amount of the different types is gigantic, and the total amount of the images is not rare. Also, most of the image data are collected from white skin tone patients. Our work put forward a new model that could improve these situations. The most significant contribution of our work is that it let the data not only focus on one type of color skin tone but is able to generate around 42 different skin tones.