Computer vision

Programme theme

Computer vision is a rapidly developing inter-disciplinary subject. It aims to tackle practical problems in several areas: visual navigation of mobile robots and road vehicles, hand-eye coordination in assembly robots, quality inspection of manufactured goods and automated analysis of medical images.

The involvement of mathematicians and statisticians has been relatively recent but substantial contributions have already been made, and there is enthusiasm in the engineering community for the benefits of a larger infusion of mathematical ideas and methods. Sufficient computing power is now available to enable theoretical approaches to be developed into feasible computational algorithms, and theoretically founded methods can be expected to have increasing importance in the future. Some recent highlights are the recovery of geometric structure in a static scene from visual information from a moving observer, the use of Bayesian methods to represent prior information on structures, and variational methods in developing systems for surface reconstruction and for dynamic feature tracking.

This programme is multi-disciplinary, with a core of mathematicians, statisticians and engineers but also including computer scientists, physiologists, psychologists, and others. The choice of themes on which the programme will focus has been driven by what are judged to be timely computer vision problems in which mathematics is likely to make an important contribution.

The computer vision programme will be carried on in three ways. First, there will be a series of special weeks, in each of which one or two lead speakers/organizers will present a particular topic, followed by seminars by participants involved in these topics developing their ideas. Second, long-term visitors will conduct their research and exchange their ideas informally and in seminars throughout the programme. Third, a series of `software challenges' will be organized (see below). In addition, four of the special weeks are marked in the programme below as Workshops. These aim to be especially attractive to short-term visitors to the Institute from both industry and academia and will present particular topics to wider audiences. The EEC Human Capital and Mobility Programme Conference Fund is being approached for support of the Workshop ``Statistical Methods in Vision'' and funding from ESPRIT and the NSF is anticipated for the Workshop in ``Geometry-driven diffusion and Image Segmentation'', which is being conducted jointly with a European/US Collaborative Working Group. People interested in attending one of the Workshops are invited to contact the Newton Institute directly (see contact below).

Schedule of Special Weeks and their Topics

• July 5-9 Active Vision I&Roger Brockett, Chris Brown
• July 12-16 Computational Geometry and Robotic Path-planning Mike Brady, Shankar Sastry
• July 19-23 Active Vision II - WORKSHOP Andrew Blake, Demetri Terzopoulos
• July 26-30 Computational Approaches to Analysis of Multiple views Olivier Faugeras, Thomas Huang
• Aug. 2-6 Psychophysics of Motion and Binocular stereo&Ken Nakayama, Brian Rogers
• Aug. 16-20 Biology of Vision David van Essen
• Aug. 23-27 Object Recognition I -- WORKSHOP Joe Mundy, Shimon Ullman
• Sept. 6-10 Object Recognition II Stuart Geman
• Sept. 13-17 Surface Geometry Jan Koenderink
• Sept. 20-24 Wavelets and Pyramid Architectures&Peter Burt, Stephane Mallat
• Sept. 27-Oct.8 STUDY BREAK
• Oct. 18-22 Statistical Methods in Vision -- WORKSHOP Peter Green
• Oct. 25-29 Speech Recognition and relations to Vision Peter Brown, John Bridle
• Nov. 1-5 Statistical Basis of Learning and Classification Brian Ripley
• Nov. 8-12 Reinforcement Learning and Adaptive Control Andrew Barto
• Nov. 22-26 Texture Segmentation and Classification Jitendra
• Malik, Knut Conradsen
• Nov. 29-Dec.3 Geometry-driven diffusion and image-segmentation -
• WORKSHOP David Mumford

Software Challenges

Computing facilities (SUN, Macintosh) are available at the Newton Institute, and vision facilities will be brought in, including image databases and framestore systems. Four software tasks are defined below as areas in which comparison between existing software systems is invited. Participants will be encouraged to bring and develop their own systems during the course of the programme, creating test suites of images on which diverse algorithms can be easily compared.

Task and Organizer(s) Texture Classification and Segmentation, Brian Ripley and Knut Conradsen, Stereo Correspondence, John Porrill, Real-time Visual Tracking, Andrew Blake, Model Based Recognition, Andrew Zisserman