Infotech Oulu Doctoral Program
Graph cut based optimisation for computer vision
Lecturer: Dr. Lubor Ladicky, University of Oxford, UK
Date, time & room
Tuesday, November 27, 2012 TS101, 12:15-15:00
Wednesday, November 28, KO102, 12:15-15:00
Thursday, November 29, TS101 14:15:-16:00
Many of the problems in computer vision can be formulated as finding the max-a-posteriori (MAP) labelling of Random Field models. In these problems each pixels in an image or in a video may be associated with a unique random variable leading to very large optimisation problems which can be efficiently solved using graph-cut based methods. In the first part a
detailed walk-through of the basic max-flow / min-cut (graph cut) algorithms will be provided, and their direct application to problems of binary segmentation will be shown. In the second, it will be shown how several classes of computer vision problems can be transformed into an equivalent graph cut problem or into a series of such problems leading to an approximate solution. In the third part recent advances in this field will be presented, as the graph constructions for several higher order and hierarchical energies will be derived and applied for object-class segmentation and stereo reconstruction problem. The course will be self-contained with the first two parts highly suitable for beginners. Advanced researchers should benefit from the final section. We shall extensively cover all important aspects of the graph-cut methods from the basics up to the current state-of-the art.
Lubor Ladicky received his master degree in Automation in 2005 at the Slovak University of Technology in Bratislava and another master degree same year in Theoretical Physics at the Comenius University in Bratislava. After a year in the industry he started his PhD in Computer Vision at the Oxford Brookes University supervised by Phil Torr and defended his thesis
in 2011. During his study he received the best paper awards at ECCV'10 and BMVC'10. At the moment I am a post-doc at the University of Oxford in the visual geometry group of Andrew Zisserman.
More information: Esa Rahtu
Last updated: 21.11.2012