CCIS Seminar: Robust Fitting and Model Selection in Computer Vision

Dear Colleagues,
Centre for Computing and Intelligent Systems (CCIS), Universiti Tunku Abdul Rahman (UTAR) will be organizing the following open seminar:

Date: 7 Oct 2011 (Friday)
Time: 11:30am-12.00pm
Venue: 6th Floor Meeting room, SA Block, FES,  UTAR, KL Campus, Jalan Genting Kelang, 53300 Kuala Lumpur
Speaker: Tat-Jun Chin
Title: Robust Fitting and Model Selection in Computer Vision
Abstract: One of the earliest examples of robust model fitting in Computer Vision is to detect lines and curves in images using the Hough Transform (Duda and Hart, 1972). Since then robust model fitting techniques have become part of the standard toolbox for Computer Vision researchers and practitioners, especially on geometry estimation problems where approaches like RANSAC dominate. Robustness is crucial in many cases as the data acquisition and preprocessing pipeline is inevitably imperfect, and this introduces outliers and noise into the data. A second important aspect is model selection, especially in the case of multiple structures, i.e. how many lines do we have in the image? Traditional model selection techniques dictate that the best model is the one that balances goodness of fit and model complexity. In this talk I will discuss ongoing work at The University of Adelaide to find new paradigms for robust model fitting and model selection. In particular I will report recently proposed methods that cast robust model fitting and model selection as statistical learning problems. These new methods offer the promise of more robustness and applicability to a wider range of problems.
Speaker Bio:
Dr Tat-Jun Chin is a Lecturer in the School of Computer Science at The University of Adelaide. He received his Ph.D. in Computer Systems Engineering from Monash University, and B.Eng in Mechatronics from the Universiti Teknologi Malaysia (UTM). Dr Chin's current research focusses on robust model fitting approaches for Computer Vision applications. In particular he is interested in searching for new paradigms for the fundamental underpinning of robust model fitting approaches, and to extend the applicability of robust model fitting approaches to a wider range of problems and operating conditions.


The seminar is open to all staff, students and public. Admission is free and no registration required. Please encourage your students / research counterparts to attend this seminar.