Abstract
This thesis presents a general approach to modeling a complex 3D scene with impossible figures that supports real-time modeling and rendering at unprecedented range of viewpoints. To achieve the connection illusion perceived when the viewer attempts to mentally glide on the surface of the impossible figure, we propose the view-dependent Laplacian deformation (VDLD): the camera guides the deformation algorithm running at high frame rate subject to connection constraints at rendering viewpoints while preserving the rigid look of impossible figures. The depth illusion is achieved by shifting and scaling the 3D model. We demonstrate complex 3D artworks and scenes for games.