Syntactic description of images took off in the late 1960s due to the seminal work done by R. Narasimhan, K. S. Fu, A. Rosenfeld, and their collaborators. The notion of stochastic image grammars has come a long way from assigning probabilities to production rules, in the early work, to a more general notion that encompasses hierarchical representations of objects and events, semantic and spatial-temporal contexts, taxonomy of visual categories, and their associated learning and inference algorithms. Recent work shows that the virtue of image grammars lies in their expressive power to represent an exponentially large number of object and event configurations by using a relatively much smaller vocabulary, and a few compositional rules. This, in turn, enables rich image interpretations.
Statistics and machine learning experience a resurgence of stochastic grammars, such as AND-OR models, deep learning, structured prediction, and Markov Logic Networks. To popularize these advances in computer vision, we organized SIG-09: 1st International Workshop on Stochastic Image Grammars, at CVPR 2009. SIG-09 hosted six keynote speakers whose talks were aimed at reducing the historical disconnect from early work, and illuminating directions for future research. Also, the workshop presented a number of peer-reviewed papers. SIG-09 was widely recognized as a successful workshop that brought together more than 200 attendees from different subcommunities. The workshop's best papers will soon be published in the IJCV's special issue.
After two years from the first workshop, we believe it is the right time and place to organize SIG-11: 2nd International Workshop on Stochastic Image Grammars, at ICCV 2011. In the last two years, there has been a tremendous progress in grammar-based formulations, specifically in terms of integrating shape, appearance, and compositionality. However, while this momentum is present in 2D object recognition, other areas of computer vision, such as 3D structure from motion, and activity recognition, seem oblivious to many empirical and theoretical advantages of probabilistic grammars. SIG-11 will be aimed at promoting grammars in a wider range of vision problems beyond 2D object recognition.