Realistic hair modeling from a hybrid orientation field

Image-based hair modeling methods enable artists to produce abundant 3D hair models. However, the reconstructed hair models could not preserve the structural details, such as uniformly distributed hair roots, interior strands growing in line with real distribution and exterior strands similar to images. In this paper, we propose a novel approach to construct a realistic 3D hair model from a hybrid orientation field. Our hybrid orientation field is generated from four fields. The first field makes the surface structure of a hairstyle be similar to the input images as much as possible. The second field makes the hair roots and interior hair strands be consistent with actual distribution. The tracing hair strands can be confined to the hair volume according to the third field. And the fourth field makes the growing direction of one point at a strand be compatible with its predecessor. To generate these fields, we construct high-confidence 3D strand segments from the orientation field of point cloud and 2D traced strands. Hair strands automatically grow from uniformly distributed hair roots according to the hybrid orientation field. We use energy minimization strategy to optimize the entire 3D hair model. We demonstrate that our approach can preserve structural details of 3D hair models.