Peter Meer (Rutgers University)
Wednesday, 11.11.2015, 10:30
In these three lectures a (personal) view is given on how from linear/nonlinear objective functions at the input, one can built non robust or robust estimation. If a higher dimensional linear space is considered for nonlinear inputs, one could, directly in total least squares; or iteratively in Levenberg-Marquardt algorithm, non-robustly estimate the features. A robust method of mean shift is presented both for segmentation and tracking. Generalizations, to nonlinear mean shift in Riemannian manifolds and to mean shift clustering in the kernel space, will be also described. Using the higher dimensional linear space, a new algorithm can be designed which, without the scales provided, returns all the inlier structures.
Peter Meer received the Dipl. Eng. degree from the Bucharest Polytechnic Institute, Romania in 1971, and D.Sc. from the Technion In 1986, both in electrical engineering .In 1991 he joined the Department of Electrical and Computer Engineering, Rutgers University, and is currently a Professor. He has held visiting appointments in Japan, Korea, Sweden, Israel and France, He was an Associate Editor of the IEEE Transaction on Pattern Analysis and Machine Intelligence between 1998 and 2002, was a Guest Editor of Computer Vision and Image Understanding for a special issue on robustness in computer vision in 2000, and was a member of the Editorial Board of Pattern Recognition between 1989 and 2005. With coauthors Dorin Comaniciu and Visvanathan Ramesh he received at the 2010 CVPR the Longuet-Higgins prize for fundamental contributions in computer vision in the past ten years. His research interest is in application of modern statistical methods to image understanding problems. He is an IEEE Fellow.