A Classified Method of Human Hair for Hair Sketching
Feng Min, Kun Zeng, Nong Sang · 2008
Human hair has significant effect on the life-likeness of human portrait and human recognition. In this paper, we present a classified method of human hair for hair sketching. We extract shape and appearance features from the training data of hair, including hair raw images and their corresponding sketching templates. Based on these features, we learn twenty-four hairstyles. Given a human hair raw image, we extract its shape and appearance features and find the best matched hair style and sketching template by Nearest Neighbor from twenty-four hairstyles. Taking the template as prototype, a new hair sketching corresponding to the raw image can be generalized by Thin Plate Spline. We test our algorithm to a large data set of hair images with diverse hair styles, experimental results demonstrate the effectiveness of our method.