Automatic Hair Colorization Using Chromaticity Distribution Matching

Human hair colorization with respect to the given model hair image without changing neither hairstyle nor hair texture is a challenging task. The principle problem making this task complicated is the difference in the texture and the illumination between a user and model images. The proposed approach is based on the matching between the user and the model hair chromaticity distributions. The idea behind the approach is to decompose the model's hair chromaticity distribution into a mix of chromaticity clusters or hair swatches. These swatches, having in general case non Gaussian color distribution are applied to the user hair. Experimental results show that proposed technique achieves high photorealistic perception and reasonable compression ratio.