Hairstyle Suggestion Using Statistical Learning

Hairstyle is one of the most important features people use to characterize one’s appearance. Whether a hairstyle is suitable or not is said to be closely related to one’s facial shape. This paper proposes a new technique for automatically retrieving a suitable hairstyle from a collection of hairstyle examples through learning the relationship between facial shapes and suitable hairstyles. A method of hair-face image composition utilizing modern matting technique was also developed to synthesize realistic hairstyle images. The effectiveness of the proposed technique was validated through evaluation experiments.