Template-based Augmentation of Visual Effects to Human Actions in Videos

MPhil Thesis Defence


Title: "Template-based Augmentation of Visual Effects to Human Actions in
Videos"

By

Mr. Xuren ZHOU


Abstract

Visual Effect (VFX) is a popular technique to enrich video content. There exist 
several light-weight mobile applications for visual effects superposition, in 
which novice users need to select a template and follow the instructions to 
create their own visual effect video. However, these applications cannot 
automatically detect the desired timing and position of the visual effect in 
the user video, thus users need to tediously adjust or add the visual effect by 
themselves.

In this thesis, we develop a fully automatic template-based augmentation system 
to superimpose visual effects onto user videos. In our system, pre-defined 
template videos with visual effects are provided and the user selects one to 
mimic the action therein. Our system processes the user imitation video and 
automatically transfers the template visual effect onto it. To this end, we 
pre- analyze the template video to detect motion interest points, one of which 
is pre-defined as the spatio-temporal location of the visual effect. Given the 
user imitation video as input, we detect its motion interest points and then 
use a dynamic programming-based algorithm to match them with the selected 
template. With the matching result, our system automatically transfers the 
visual effect from the template to the desired spatio-temporal location in the 
user video. We conduct a user study to demonstrate the effectiveness of our 
system.


Date:			Tuesday, 26 June 2018

Time:			10:00am - 12:00noon

Venue:			Room 5560
 			Lifts 27/28

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


**** ALL are Welcome ****