Realistic human head modeling with multi-view hairstyle reconstruction

We present a method for constructing photorealistic 3D head models from color images and a geometric head model of a specific person. With a simple experimental setup, we employ a user-assisted technique to register the uncalibrated images with the geometric model. A weighted averaging method is then used to extract a panoramic texture map from the input images. To recover the hairstyle of the specified person, a virtual photo plane is defined, according to a corresponding true photo plane, on which one input image is recorded. It provides hints to compute the 3D positions of the visual contour points of the hair from the images taken at different viewpoints. Finally, more hairs are grown to cover the whole region of the scalp using an interpolation method on the 3D scalp mesh surfaces.