IEEE WORKSHOP on REPRESENTATIONS OF VISUAL SCENES (In Conjunction with ICCV) Organizers: P. Anandan, David Sarnoff Research Center, anandan@sarnoff.com Harpreet Sawhney, IBM Almaden Res. Ctr., sawhney@almaden.ibm.com Amnon Shashua, Hebrew Univ., Israel, samm@cs.huji.ac.il Eero Simoncelli, Univ. of Pennsylvania, eero@tarpon.cis.upenn.edu General Chair: Takeo Kanade, CMU Program Committee: Ted Adelson, MIT Richard Hartley, GE Steve Maybank, Oxford John Oliensis, NEC Roz Picard, MIT Carlo Tomasi, Stanford Thierry Vieville, INRIA Andrew Zisserman, Oxford The purpose of this workshop is to bring together researchers interested in representation of visual scenes based on a collection of images (either video or a collection of stills). In recent years, two clear trends have emerged in the area of multiple image analysis: image-to-image alignment based on motion-field models, and recovery of 3D geometry in terms of affine and projective invariants. Work in the first area has led to techniques for panoramic view construction based on multiple-image alignment and layered representations, whereas work in the second area has lead to techniques for the representation of scenes as collection of views and the recovery of a new view as combinations of a set of given views. Taken together, the new trends have led to new ways to formulate the traditional problems of structure-from-motion and 3D scene reconstruction. These emerging techniques are also closely related to work in active/animate vision, especially to methods that use fixation and tracking. Their benefits to visualization, video databases, compression, and handling uncalibrated imaging situations is already becoming apparent. However, one of the major challenges, and a recent topic of lively interest, is to obtain an adequate generalization of the fundamental structures associated with two views to multiple-view situations. Such generalizations are essential for understanding view-based representations, visual database indexing, numerical stability of 3D recovery, and navigation. Several recent results suggest that some of the accuracy and stability problems that have plagued the previous generation of techniques can be overcome. Furthermore, the issues of sparse versus dense scene representations, and their usefulness for recognition, navigation and intelligent video manipulation are areas of active research. The workshop will be held on 24 June, the day after the end of ICCV95. Hotel rates for ICCV95 have been extended. It will include 12-15 contributed presentations and a panel discussion at the end of the day. Papers that are submitted will be reviewed by the program committee. A proceedings will be made available a few months after the workshop. Prospective authors are invited to submit papers on topics described above. Submissions should be 5-10 pages, including figures and references. Send four copies of the paper summary and a cover sheet stating the (1) paper title, (2) Brief 2-3 sentence summary of the topic and contribution, (3) contact author's name, (4) address, (5) telephone number, (6) Fax number, and (7) electronic mail address to: Dr. P. Anandan David Sarnoff Research Center CN 5300 Princeton, N.J. 08543 Important Dates: Deadline for Submission: 25 March 1995 Notification of Acceptance: 1 May 1995 Camera-Ready papers (10pp) due: 1 June 1995