This talk is devoted to the problem of dynamic texture (DT) classification using optic flow features. We provide a principled analysis of local image distortions and their relation to optic flow, then propose a framework for quantitative temporal periodicity analysis of DTs. Adapting the SVD-based algorithm for signal period estimation by Kanjilal et al.(1999), we measure the degree of periodicity of natural dynamic textures. Then we present the results of a comprehensive DT classification study that compares the performances of different flow features for a normal flow algorithm and four different complete flow algorithms. The efficiencies of two flow confidence measures are also studied. Finally, we demonstrate that adding the optic flow based temporal periodicity features improves the classification rates.

The present talk aims at explaining the physical phenomena governing the interaction of light, paper and ink halftones and at presenting classical spectral reflectance prediction models. Physical phenomena comprise surface reflections and refractions at the air-paper interface, the propagation of light within the paper, internal reflections at the paper-air interface, ink spreading and trapping. We review classical reflectance prediction models such as the spectral Neugebauer model, the Yule-Nielsen modified spectral Neugebauer model, and the multiple reflection Clapper-Yule model. We then discuss the phenomena of dot gain and ink spreading and show how to take them into account. Finally, we give an overview of recent progress in the field and point to yet unsolved problems.

Algorithms for MAP estimation in MRFs (or, alternatively, for minimizing energy functions of discrete variables) are now routinely used in computer vision. Minimization techniques based on graph cuts are probably the most popular ones; they often outperform other algorithms and give state-of-the art results for many problems.

In this talk I will concentrate on an alternative minimization algorithm tree-reweighted max-product message passing (TRW) introduced recently by Wainwright et al. I will present some new developments which show that TRW is a serious competitor to graph cuts.

The algorithm of Wainwright et al. gives strong results in practice, but unfortunately it is not guaranteed to converge. I will present a new algorithm called TRW-S which is related to TRW but has guaranteed convergence properties. Experimental results demonstrate that on a real stereo problem TRW-S significantly outperforms the algorithm of Wainwright et al. and obtains lower energy than graph cuts and max-product belief propagation.

In this talk, I will present a generic method for solving Markov random fields (MRF) by formulating the problem of MAP estimation as 0-1 quadratic programming (QP). Though in general solving MRFs is NP-hard, we propose a second order cone programming relaxation scheme which solves a closely related (convex) approximation. In terms of computational efficiency, our method significantly outperforms the semidefinite relaxations previously used whilst providing equally (or even more) accurate results.

Unlike popular inference schemes such as Belief Propagation and Graph Cuts, convergence is guaranteed within a small number of iterations. Furthermore, we also present a method for greatly reducing the runtime and increasing the accuracy of our approach for a large and useful class of MRF. We compare our approach with the state-of-the-art methods for subgraph matching and object recognition and demonstrate significant improvements.

Joint work with Philip Torr and Andrew Zisserman.

The core part of our approach is a flexible formulation for object shape that can combine the information observed on different training examples in a probabilistic extension of the Generalized Hough Transform. The resulting approach can detect categorical objects in novel images and automatically infer a top-down segmentation from the recognition result. The segmentation is then used to again improve recognition by allowing the system to focus on object pixels and discard misleading influences from the background. Moreover, the information from where in the image a hypothesis draws its support is used in an MDL based verification stage to resolve ambiguities between overlapping hypotheses and factor out the effects of partial occlusion.

As an application, we address the problem of detecting objects such as cars, motorbikes, and pedestrians in real-world street scenes. Qualitative and quantitative results on several challenging data sets confirm that our method is able to reliably detect objects in crowded scenes, even when they overlap and partially occlude each other. In addition, the flexible nature of our approach allows it to operate on very small training sets.

We approach the corresponding optimization problem by the TRW-S (sequential Tree-reweighted message passing) algorithm [Wainwright2003, Kolmogorov2005]. Our model design allows for a considerably wider class of smooth transformations and yields a compact representation of the optimization task. For this model, the TRW-S algorithm demonstrated nice practical performance in experiments.

We also propose a concise derivation of the TRW-S algorithm as a sequential maximization of the lower bound on the energy function.

Joint work with Ivan Kovtun and Vaclav Hlavac.

In this talk I'll describe how a graph structure can be built and used to describe the interactions that occur. Given this platform, I'll present two separate strategies for labelling the identity of the nodes of this graph. Results will be shown from an international football match captured by a multi-camera rig that produces a wide screen video that is full stationary view of the pitch.