Home Call for papers People Dates Program Submission CVPR'10

9:00 - 10:00 Keynote talk: Matching images with deformations,
David Jacobs (University of Maryland)

When we match images that come from the same object, we must often allow for 2D, non-linear deformations. These can model changes in shape that can occur when an object deforms or an articulated object moves its parts, differences in shape between different instances of the same type of object, or variations in apparent shape due to changes in viewpoint. This talk will provide an overview of several approaches to matching that stress finding problem formulations that yield computationally efficient algorithms. First, I will present an approximation algorithm for computing the Earth Mover's Distance (EMD), a metric for comparing probability distributions that can be used to match image descriptors, accounting for deformations. Using a wavelet-based representation, we construct an accurate, linear time algorithm for computing the EMD. I will then describe a novel image descriptor that is invariant to deformations, and a shape descriptor that is invariant to articulations. We use this shape matching algorithm in a hand-held device for computer assisted plant species identification. I will then show that the cost returned by a stereo matching algorithm can be used for image matching when there deformation due to pose variation. We use this to construct a face recognition algorithm that compares 2D gallery and probe images taken from different viewpoints. This algorithm outperforms all prior work on the CMU PIE data set for face recognition with pose variation.

10:00 - 10:30 Coffee break

2D shapes and deformable images
  • 10:30 - 11:00 Content-aware image resizing by quadratic programming, Renjie Chen, Daniel Freedman, Zachi Karni, Craig Gotsman, Ligang Liu
  • 11:00 - 11:30 Local pose estimation from a single keypoint Alberto Del Bimbo, Fernando Franco, Federico Pernici
  • 11:30 - 12:00 Straight skeletons for binary shapes Markus Demuth, Franz Aurenhammer, Axel Pinz

13:30 - 14:30 Keynote talk: On surface comparison and symmetry,
Yaron Lipman (Princeton University)

In the first part of the talk we will suggest a method for automatic surface comparison and alignment based on principles from conformal geometry and optimal mass transportation. One application of the method is automatic calculation of point correspondences between surfaces. Another application is a novel distance definition between disc-type surfaces. In the second part of the talk, we will discuss the relation between symmetry and the shape matching problem. We will present a recent result of how symmetry can be understood and/or used in this context.

Shape representation and inverse problems
  • 14:30 - 15:00 Continuous Procrustes analysis to learn 2D shape models from 3D objects Laura Igual, Fernando De la Torre
  • 15:00 - 15:30 Persistence-based segmentation of deformable shapes, Primoz Skraba, Maks Ovsjanikov, Frederic Chazal, Leonidas Guibas

15:30 - 16:00 Coffee break

16:00 - 17:00 Keynote talk: Metric geometry in shape matching,
Facundo Mémoli (Stanford University)

The problem of object matching under invariances can be studied using certain tools from Metric Geometry. The central idea is to regard objects as metric spaces (or measure metric spaces). The type of invariance one wishes to have in the matching is encoded by the choice of the metrics with which one endow the objects. The standard example is matching objects in Euclidean space under rigid isometries: in this situation one would endow the objects with the Euclidean metric. More general scenarios are possible in which the desired invariance cannot be reflected by the preservation of an ambient space metric. Several ideas due to M. Gromov are useful for approaching this problem. The Gromov-Hausdorff distance is a natural first candidate for doing this. However, this metric leads to very hard combinatorial optimization problems and it is difficult to relate to previously reported practical approaches to the problem of object matching. I will discuss different adaptations of these ideas, and in particular will show a construction of an L^p version of the Gromov-Hausdorff metric called Gromov-Wassestein distance which is based on mass transportation ideas. This new metric leads directly to quadratic optimization problems on continuous variables with linear constraints. As a consequence of establishing several lower bounds, it turns out that several invariants of metric measure spaces are quantitatively stable in the GW sense. These invariants provide practical tools for the discrimination of shapes and connect the GW ideas to several pre-existsting approaches. After reviewing these constructions I will explain more recent developments including spectral versions of the GW distance and connections with persistent topology invariants.

Shape similarity
  • 17:00 - 17:30 Bypass information-theoretic shape similarity from non-rigid points-based alignment Francisco Escolano, Miguel Lozano, Boyan Bonev, Pablo Suau
  • 17:30 - 18:00 Shape matching based on diffusion embedding and on mutual isometric consistency Avinash Sharma, Radu Horaud