Many problems in computer vision, such as segmentation, stereo, and image restoration, are formulated as optimization problems involving inference of the maximum a posteriori (MAP) solution of a Markov Random Field (MRF). Such optimization schemes have become quite popular, largely owing to the success of optimization techniques such as graph cuts, belief propagation, and tree-reweighted message passing. However, because of the lack of efficient algorithms to optimize energies with higher-order interactions, most have been represented in terms of unary and pairwise clique potentials, with a few exceptions that consider triples. This limitation severely restricted the representational power of the models: The rich statistics of natural scenes cannot be captured by such limited potentials. Higher order cliques can model more complex interactions and reflect the natural statistics better. In this talk, I will present the recent advances in optimizing higher-order MRFs, especially with graph cuts.
Hiroshi Ishikawa received the BS and MS degrees in mathematics from Kyoto University and the Ph.D. degree in computer science from the Courant Institute of Mathematical Sciences, New York University. In 2004, he became an assistant professor at Nagoya City University, where he had been promoted twice by 2010, when he moved to Waseda University as a full professor. Currently, he is also a Sakigake Researcher in the JST Mathematics Program at the Japan Science and Technology Agency. He has also received the IEEE Computer Society Japan Chapter Young Author Award in 2006 and the MIRU Nagao Award in 2009.