“Hair detection in images is useful for many applications, such as face and gender recognition, video surveillance, and hair modelling.”
– prompting the development of a new Convolutional Neural Network (CNN) software application which can sort images of hair styles into straight, wavy, curly, kinky, braids, dreadlocks, and short – with an accuracy of around 90%.
The system for hair detection, segmentation, and hairstyle classification in the wild (that’s to say outside the lab) has been developed by researchers at the University of Brescia, Italy, Queen Mary University of London, and the University of Glasgow.
“Even if attitudes towards hairstyles or hair removal may vary widely across different cultures and historical periods, hairstyle is one of the defining characteristics of humans, often related to person’s social position, such as age, gender, religion. Despite the fact that human [sic] can drastically manipulate their hair, they typically do not, so that hair appearance and attributes may provide useful cues also for the recognition task. Hence, as another contribution, for the first time automatic hairstyle recognition is performed here by means of a multi-class texture classification step on the previously segmented hair region.”
See: Hair detection, segmentation, and hairstyle classification in the wild in Image and Vision Computing,Volume 71, March 2018, Pages 25-37.
BONUS The research team have created a Multi-class hair image database (straight, wavy, curly, kinky, braids, dreadlocks, and short-men) with Ground Truth, which is freely available for download here (42MB)