TITLE: Shape Reconstruction of 3D Smooth Surfaces
PRESENTER: Dr. Yael Moses
AFFILIATION: National ICT Australia, Symbolic Machine Learning and Knowledge Acquisition
DATE: Friday 8th October 2004
TIME: 12 noon - 1 pm
PLACE: CSE K17 1st Floor Seminar Room
Retrieving the three dimensional shape of the world from two dimensional images is one of the challenging problems of computer vision. Recovering the three-dimensional shape of objects or scene has many applications including object recognition, face recognition, and computer-graphic applications. In the first part of the talk I will give some background on the challenges in object recognition and 3D shape recovery. The second part of the talk will present a method for recovering the three-dimensional shape of a featureless smooth surface.
Existing approaches to shape recovery of three-dimensional objects can be classified into geometric or photometric methods. Geometric methods, namely stereo or structure from motion, are based on two or more images of the same object taken from different viewpoints. These methods are limited to images for which correspondence between image points can be defined.
Photometric methods, on the other hand, are more appropriate for recovering the 3D shape of smooth surfaces. Photometric methods use reflectance models to transform image brightness into 3D-shape. The robust photometric methods are also based on two or more images of the same object, and require correspondence between images. This is usually obtained by using a set of images that were all taken from the same viewpoint. Both photometric and geometric stereo methods require known correspondence.
The method proposed here provides dense correspondence between images of featureless surfaces by integrating the photometric and geometric information obtained from a set of calibrated images taken from different viewpoints and illuminated by different point light sources. This is joint work with Ilan Shimshoni, from the Technion Israel.
BIOGRAPHY OF SPEAKER:
Dr. Yael Moses received her Ph.D in computer science from the
Weizmann Institute of Science, Rehovot, Israel, in 1994. She spent a year in
the Engineering department at Oxford University, and four more years in the
Weizmann Institute of Science. Then she moved to the Interdisciplinary Center
Herzeliya, Israel as a senior lecturer. Currently, she is a visiting
researcher in NICTA. Her main research interests include human vision, computer vision, distributed vision systems, and applications of computer vision to multimedia systems.
Host & Seminar Convener:
National ICT Australia, Symbolic Machine Learning & Knowledge Acquisition