SIV: A Single-View Visual Modeling Tool

(click on the image for a larger view)

Abstract: Shape recovery techniques, such as shape from shading, have been the target of much research but have still not proven to be successful in general cases. They suffer from sensitivity to initial conditions, noise and often not enough constraint specifications. I propose an interactive tool, SIV, that allows the user to specify constraints (which manifest themselves as constraints on the normal vector at various regions in the image) to better guide the visual modeling process. Furthermore, shape from shading techniques often require that the object in question be lambertian. SIV offers the ability to operate on intrinsic images that have their specularities removed, thereby opening the doors to model specular objects as well. Finally, SIV also has the ability to aid in the determination of the object’s BRDF.