EECS CSPL SEMINAR SERIES
FALL TERM 1999


R. Kakarala

Dr. Ramakrishna Kakarala

Visiting Scholar to EECS Department

On sabbatical leave from

University of Auckland, New Zealand

kakarala@eecs.umich.edu



Thursday, October 14
4:30 - 5:30 P.M.
Room 1311 EECS


A Multiresolution form of the SVD and its applications to image analysis

Abstract-
In many applications (e.g, image compression) it is useful to obtain a statistical characterisation of an image, at each of several levels of resolution. For this purpose, a multiresolution form of the singular value decomposition (SVD) is proposed. It is well known that a SVD decomposes a matrix into orthogonal components with which optimal subrank approximations may be obtained. The multiresolution SVD proposed here retains those properties, and moreover, has linear computational complexity. By using the multiresolution SVD, the following important characteristics of an image may be measured, at each of several levels of resolution: isotropy, sphericity of principal components, self-similarity under scaling, and resolution of mean-squared-error into meaningful components. Theoretical calculations are provided for simple statistical models to show what might be expected. Results on real images are provided to illustrate the usefulness of the multiresolution SVD.



Biosketch-
R. Kakarala is currently a Visting Scholar with the EECS Dept, on sabbatical leave from the University of Auckland. He is a graduate of U. Michigan (B.S. 1986, M.S. 1988), and of U. California, Irvine (PhD 1992). His research interests are in image and video compression, statistical signal processing, and computer vision. His home page is here


return to CSPL Seminars homepage