Local Feature Learning and Non-Rigid Matching

Gustavo Carneiro, Tat-Jun Chin and Ian Reid


Local feature extraction and matching are fundamental tasks for various image processing and computer vision applications, such as image registration, image super-resolution, 3D reconstruction and augmented reality. Though both feature extraction and matching are much studied, in recent times new methods have emerged for doing both. In particular, there has been burgeoning interest in the use of machine learning to extract local features, instead of using off-the-shelf detectors. Popular examples of such techniques include those based on distance metric learning and deep belief learning. Tests show clear improvements over features from off-the-shelf detectors. Secondly, there is also renewed interest in the problem of non-rigid structure from motion from local features. We focus on mismatch removal in non-rigid modelling, for which no standard solution has emerged (the rigid case is well studied and numerous mature algorithms exist).

This tutorial therefore focuses on a modern look at the above areas, demonstrating through theory and examples on how the new techniques work, how to use them, and when and where they can be beneficial.


Gustavo Carneiro is a Senior Lecturer the School of Computer Science at the University of Adelaide (equivalent to associate professor in North America) in 2011. From 2008 to 2011 Dr.Carneiro was a Marie Curie IIF fellow and a visiting assistant professor at the Technical University of Lisbon (Instituto Superior Tecnico) within the Carnegie Mellon University-Portugal program (CMU-Portugal). From 2006 to 2008, Dr. Carneiro was a research scientist of the Integrated Data Systems Department at Siemens Corporate Research in Princeton, USA. In 2005, he was a post-doctoral fellow at the the University of British Columbia with Professor David Lowe and at the University of California San Diego with Professor Nuno Vasconcelos. Dr. Carneiro received his Ph.D. in computer science from the University of Toronto under the supervision of Professor Allan Jepson in 2004. Gustavo Carneiro is originally from Brazil.

Tat-Jun Chin received the BEng (Mechatronics) from Universiti Teknologi Malaysia (UTM) in 2003. In 2003-2004 he worked for Agilent Technologies' Sensor Solutions Division as a Test Engineer. In 2004, he received the Australia-Asia Award (AUD$100,000) to pursue his Ph.D. in computer vision at Monash University, Victoria, Australia. He was a Research Fellow at the Institute for Infocomm Research (I2R) in Singapore from 2007-2008. Since 2008 he was a Postdoctoral Research Fellow, then Lecturer at The University of Adelaide, South Australia. His research interests include robust estimation and statistical learning methods in Computer Vision. He has served as Program Committee member and reviewer for the major computer vision conferences and journals, and is currently the Co-Treasurer of ICIP 2013.

Ian Reid is a Professor of Computer Science at The University of Adelaide. Previously he was a Professor of Engineering Science and Fellow of Exeter College, Oxford University. Together with long-time colleague Prof. David Murray, he ran the Active Vision Group at Oxford which is part of the wider Robotics Research Group. His research interests span a wide range of topics in Computer Vision. In particular he is concerned with algorithms for visual control of active head/eye robotic platforms (for surveillance and navigation), visual geometry and camera self-calibration (applications of these to measurement, AR and VR, including sporting events), visual SLAM, human motion capture, activity analysis, and novel view synthesis.