Foundations of digital sketch processing

点击翻译按钮获取中文摘要

Sketching, or drawing of curved strokes on a flat surface, is a natural way to communicate shapes. Experts and amateurs alike ubiquitously employ sketches to depict both real and virtual content, including characters, machinery, molecules, and abstract shapes. As such sketches serve as both an art form and a medium for inter-human shape communication, and offer an avenue to communicate shapes between human users and computer systems. Existing systems for sketch processing are not capable of processing complex, imprecise sketches instead typically targeting clean drawings.The long-term goal of our research is to develop digital sketch processing algorithms that will form the foundation of future robust sketch processing systems for 2D/3D content creation. In pursuing that goal, we will take a two-pronged approach serving two targets: We will create algorithms and systems viewing 1)Sketches as an Art Form, i.e. assisting humans in improving the quality of the sketches they draw, and 2) Sketches as Shape Representation, i.e. enabling humans to use sketches for human-computer communication. 1.Sketch As Art Form. In our first line of research, we will develop algorithms assisting humans to improve their sketches, creating digital art. We will algorithmically improve clarity, aesthetics, and expressivity of human drawn content,leveraging the perceptual principles from artistic literature and human perception theories,e.g. Gestalt psychology. In particular, we will create a style-independent sketch consolidation based on sub-stroke analysis and region extraction. We will develop algorithms to colorize and shade sketches. 2.Sketch As Shape Representation. In our second line of research, we will facilitate human-computer communication via sketches,i.e. using sketches to communicate 2D/3D shapes to content creation systems. We will address the fundamental causes of instabilities of modern systems using sketches, enabling more robust systems for 2D cartoon animation and 3D content modeling via sketches. We will develop the geometry processing toolkit viewing rough sketch as an approximate depiction of the underlying geometry, similar to how point clouds or meshes approximate an underlying surface. We'll use Boundary Element Methods to discretize Laplacian operator on sketches, enabling state-of-the-art Laplacian-based methods to deform or find correspondences between geometries for future morphing and inbetweening systems. We will analyse human inaccuracies when drawing sketches and create algortihms inferring drawing topology from a rough sketch. This will enable topology-sensitive systems (e.g sketch-based modeling) to correctly interpret complex natural sketches. We'll use sketches as etchings to convey textures/shading/shadows, fur/hair, to surfaces.The PIs are uniquely positioned for this research: it requires expertise of perception-based research and optimization (Prof.Sheffer), in mathematics/discrete differential geometry (Prof.Bessmeltsev), and in rendering,shading,texturing (Prof. Poulin). Both Profs.Sheffer and Bessmeltsev have an extensive publication record in sketch processing/modeling, including 5 shared publications, Profs.Sheffer and Poulin shared 1 paper on volume modeling.Tangible outcome will be algorithms and systems in the area of applications required by Canadian, in particular Quebec, film/animation/gaming and manufacturing industries, where content creation is one of the main time-consuming and laborious tasks. We will facilitate more robust and efficient content creation systems using sketches, allowing artists and designers to focus their time on creative tasks.