Computer Sciences Seminar
Friday, April 22, 2004 

Abstract

Stereo correspondence is a fundamental technique in image analysis using which we may estimate the depth information of a 3-D object. The extraction of depth-information is done by the human visual system in a natural way but it is a challenging task for a camera system. Current methods for finding corresponding points in stereo images can be divided into two categories. In the "calibrated" methods, information about the position of the cameras and other constraints are given. In the "un-calibrated" methods, such information is not necessarily available - these methods are attractive in many situations when the calibration data of the cameras are hard to obtain or keep fixed. In this talk we shall describe an un-calibrated stereo correspondence method using wavelets. In our method, the approximation channel of the decomposition is combined with a hierarchical feature-based matching, and the results from the coarse level are propagated to finer levels. Wavelet theory has found powerful applications in many areas including visual perception, image compression, design of charge-coupled devices, fractals and turbulence.

----------------------------------------

PRABIR BHATTACHARYA obtained his Ph.D. from the University of Oxford, UK in 1979.

Till 2001, he was a Full Professor at the University of Nebraska-Lincoln, Department of Computer Science and Engineering.
He then joined the Panasonic Technologies Laboratory in Princeton, New Jersey as a Principal Scientist and Project Leader.

His research interests are on image processing and multimedia security. He is on the editorial boards of five journals including the IEEE Transactions on Systems, Man and Cybernetics. He is a Fellow of the IEEE, and a Fellow of the International Association for Pattern Recognition (IAPR).