Abstract: Hair segmentation is a big challenge in the field of image segmentation. Automatic hair segmentation is of great significance to assist gender classification, identity recognition, medical image analysis, head reconstruction, AR hair dyeing and so on. Automatic hair segmentation based on machine learning are common methods in this field, which have the advantages of high efficiency and good performance. This paper combs the development history of traditional automatic hair segmentation based on early machine learning and deep learning, and focuses on the traditional segmentation methods such as Bayesian model, region growth algorithm, clustering algorithm and graph cutting algorithm, as well as the segmentation methods based on deep learning such as CNN, FCN, U-Net and Mobilenet. The segmentation effect, advantages, disadvantages and development direction of each method are summarized and compared. The hair segmentation method based on deep learning needs to use a large amount of data sets to train the network. This paper sorts out the attributes of the commonly public data sets for hair segmentation, and compares the segmentation performance of different data sets. On this basis, this paper combs and analyzes the difficulties and challenges faced by automatic hair segmentation based on machine learning, puts forward solutions to the existing problems, and looks forward to the development prospect of this field.