Semantic-Based Interactive Shape Analysis and Manipulation

PhD Thesis Proposal Defence


TITLE: "Semantic-Based Interactive Shape Analysis and Manipulation"

by

Mr. Youyi Zheng


ABSTRACT:

Geometric modeling is a fundamental problem in computer graphics. The 
exponential growth of geometric data on the internet has shifted 
researchers' focus from computing local, low-level geometry features such 
as curvatures and textures to high-level semantic information such as 
shape parts information and shape structural characteristics (symmetry, 
parallelism, etc.). However, automatically computing these semantic 
information is difficult. It is an ill-posed problem since shape semantics 
is based on human perception, which may be subjective even towards the 
same object.

This thesis focuses on developing interactive approaches that exploit 
shape semantic information, and respect and preserve them during 
manipulation and editing. We leverage the user assistance to fully reveal 
and respect semantic information during geometric modeling processes. In 
particular, this thesis aims at advancing the state-of-the-art techniques 
by incorporating the user intent in two specific geometric applications: 
shape segmentation and shape manipulation.

First, we introduce two interactive tools for shape segmentation, which we 
call cross-boundary brushes and dot scissor. Both tools offer very simple 
and easy-to-use user interfaces that operate at interactive rates. In 
contrast to the state-of-the-art interactive segmentation tools, our tools 
allow the user to cut out meaningful and functional components using only 
a single mouse stroke or click near boundary regions in most cases, making 
them very convenient to use. We adopt the concept of isolines of harmonic 
fields as cutting boundaries in designing both tools. We show that the 
propagation properties and the differentiating power of the harmonic 
fields allow effective computation of shape semantic boundaries for 
segmentation purpose.

Second, we developed an editing framework that first extracts the shape 
structural features and preserves them during user manipulation. In 
contrast to traditional shape editing framework, the system operates at 
the component level and takes shape structural characteristics such as 
inter-relations among semantic components as modeling constraints, 
enabling an effective structure-preserving editing tool. We show that user 
assistance is essential in accurately revealing the complex shape 
structures. We use a semi-automatic shape segmentation process as a prior 
step to facilitate the analysis of shape structures and inter-relations, 
and show that these shape analysis results play an important role in 
preserving the shape global features during user manipulation.

Date:                   Wednesday, 23 February 2011

Time:                   3:00pm - 5:00pm

Venue:                  Room 3401
                        lifts 17/18

Committee Members:      Dr. Chiew-lan Tai (Supervisor)
                        Prof. Long Quan (Chairperson)
                        Dr. Huamin Qu
                        Dr. Pedro Sander


**** ALL are Welcome ****