Dynamic Sketching with Structure-aware Shape Analysis

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Dynamic Sketching with Structure-aware Shape Analysis"

By

Mr. Jingbo LIU


Abstract

Line drawings can be remarkably efficient at conveying shape and meaning while 
reducing visual clutter. Inspired by the effectiveness and aesthetic appeal of 
human line drawings, researchers have investigated algorithms for generating 
line drawings from 3D meshes. Almost all such techniques focus on only the end 
product; very few regard the line drawings as a creative process. The creation 
process of a drawing provides a vivid visual progression, allowing the audience 
to better comprehend the drawing. It also enables numerous stroke-based 
rendering techniques.

In this dissertation, we address the problem of simulating the process of 
observational drawing; that is, what and how people draw when sketching. Given 
a 3D model and a viewpoint, our method synthesizes a visually plausible 
simulation of an observational sketching process. To conveniently change the 
view, we design a novel touch-based interface that supports six degrees of 
freedom 3D direct manipulation while requiring only two-finger operations.

We develop structure-aware shape analysis methods to obtain the intended 
drawing trajectories for organic objects and urban scenes separately, which 
address the question of what do people draw. Apart from the trajectories that 
depict visual features using conventional local geometric properties, we focus 
on the auxiliary trajectories indicating the composition of the drawing. We 
extract the auxiliary trajectories from contextual properties such as the 
topological layout, proportions of object parts, fitted primitives, partial 
symmetries, and levels of abstractions.

We develop the humanized stroke synthesis and stroke ordering methods to 
address the question of how do people draw. The stroke synthesis method 
simulates the action of a human moving a pen along an intended trajectory using 
a feedback control system. It produces human-like tentative strokes with 
inexact tracing and retracing effects. To assign a drawing order to the 
strokes, we view the sketching process as analogous to an information delivery 
process. A novel concept of sketching entropy, which measures the shape 
information of a stroke, is introduced. We obtain the complete drawing order by 
requiring every next drawn stroke to maximize the information gain. Finally, we 
use the humanized strokes and their ordering to create the sketching animation.

We conduct a user study to evaluate the visual plausibility of the simulated 
drawing processes and the effectiveness of our proposed method. Experiment 
confirms that our results are visually plausible. The statistical analysis 
shows that our entropy-based ordering strategy leads to more plausible results 
than those driven by the conventional Gestalt rules used in previous works.


Date:			Monday, 24 August 2015

Time:			10:00am - 12:00noon

Venue:			Room 2131C
 			Lift 19

Chairman:		Prof. Robert Ko (LIFS)

Committee Members:	Prof. Chiew-Lan Tai (Supervisor)
 			Prof. Chi-Shing Chung
 			Prof. Long Quan
 			Prof. Shuhuai Yao (MAE)
 			Prof. Pheng-Ann Heng (Comp. Sci. & Engg., CUHK)


**** ALL are Welcome ****