5. towards Inverse Bcsdf Rendering of Human Hair

HAIR In the previous chapters a frame work for synthesis of photo-realistic hair images has been presented. In this chapter a discussion of the inverse problem will be made: Assuming that photographs of a hairstyle and the hair geometry are given, we are addressing the estimation of the BCSDF of the hair fibers. Such an image-based reconstruction can be valuable for many areas of application. For instance the " hair color " of a real actor could be analyzed and transferred to a virtual character or the effects of cosmetic styling products (such as increased shininess) could be evaluated automatically. The classification of a " hair color " could also be enhanced. Since the appearance of hair strongly depends on lighting and hairstyle, substantial expert knowledge is needed to classify a hair color by hand (Fig.5.3). By using an image-based classification an even a much more subtle distinction would be possible. In chapter 3 light scattering from fibers was discussed and a flexible physically-based BFSDF (inducing a BCSDF) for dielectric fibers was presented. It was shown that such a scattering function is a suitable basis to approximate scattering from human hair. However, accurately fitting this function to real hair would require costly fiber based measurements. On the other hand, for many areas of application—such as the classification of a hair color—an accurate fiber based reconstruction of the BFSDF is not essential. Hence, if the main focus is not on accuracy, an image based approach roughly estimating the BCSDF can be a viable option. In order to make this problem feasible we extensively use the knowledge on the specific optical properties of human hair. Note that the ideas sketched in the following sections, as well as our " proof-of-concept " implementation, should be seen only as a starting point for further substantial research.