Motion Estimation using Lie Operators

Dr. David Pan

Department of Electrical and Computer Engineering
University of Alabama in Huntsville

April 9, 2004
202 Madison Hall
3:00 PM (Coffee and Cookies at 2:30)

Abstract

Motion estimation is a critical component in numerous international standards for video coding. Motion estimation with increased accuracy can better reveal the temporal redundancy existing in video sequences and improve the coding efficiency. Conventional motion estimation techniques in video coding rely on displacement motion vectors and thus have limitations in capturing non-translational transformations such as scaling, rotations and deformations of objects in video sequences. We find the theory of Lie groups and Lie algebra useful as it enables us to establish a formal connection between groups of transformation of an input space (such as the translation, scaling, and etc.) and their effect on a functional of that space. In this talk, we will briefly introduce the technique of Lie operators, which are linear operators with low complexity when implemented in video coding systems. We will show that Lie operators are capable of detecting small degrees of object transformations and hence they can be used to achieve more accurate motion estimation. We will also explore the interesting interactions among the accuracy of motion estimation, side information for characterizing transformation parameters, and the computational complexity associated with several novel motion estimation algorithms based on the Lie operators.