Coherent 3D line drawings for stylized animation

Creating hand-drawn animation is traditionally a laborious task, in which artists are highly involved to draw key frames and, thereafter, create inbetween frames from each pair of key frames. Although the development of paperless animation systems and auto-inbetween techniques somehow relieves artists' burden of creating many inbetween frames by hand, artists still have to painstakingly do all key-frame drawings by hand or using a computer. An alternate approach to avoid the laborious manual drawing task is to incorporate automatic feature line extraction from animated 3D models to create 2D animations. Computer-generated lines could be used as initial drawings for artists, where they can further refine or exaggerate lines with personal styles for animation production. %We aim to free artists from the laborious drawings of basic shapes and let them focus more on the part of creative work. This thesis proposes a solution consisting of several techniques and an integrated framework for automatic line extraction in 2D animation. The first technique is a 3D line extraction method to extract lines from 3D triangle mesh models to closely resemble what artists would perceive and draw. The second technique is a method to generate temporally coherent 3D lines from 3D models using innovative hysteresis thresholding. The third technique is a matching algorithm to create correspondence of 3D individual lines (i.e., paths) across a sequence of frames. The second and the third techniques are used for sequencing the extracted lines frame-by-frame into an animation for temporal coherence. Each of the three algorithms is a valued contribution to Non Photorealistic Rendering. Finally, an integrated framework is proposed with a novel interactive tool set for artists to create 2D stylized animations by incorporating automatic line extraction from animated 3D models. The framework is artist driven -- it allows artists to edit or stylize line work in 2D, similar to what they do in a traditional animation pipeline but utilizing 3D information. The framework is validated with a 2D animation software. We demonstrate the effectiveness of the framework in creating an animation with the software. %It is a successful attempt for these NPR algorithms to be used in industry.DOCTOR OF PHILOSOPHY (SCE