Dragomir Anguelov(drago@cs.stanford.edu)

Affine Motion-compensated Prediction 
------------------------------------

Traditional motion-compensated coders use only translational motion
estimates for motion-compensated prediction. The goal of this project is
to compare the traditional motion-compensated hybrid coder (TMCC), with
its counterpart, where we have an affine motion model instead of the
translational one (AMCC).

The affine motion model describes a plane's motion in 2D space and offers
potentially better description of block motion, especially for some camera
motions like zoom and rotation, and for complex motion of objects between
frames. The drawback of the affine model is the increase in the number of
parameters (6 parameters instead of 2 for translation) and increased
computational cost. 

I plan to implement an algorithm which computes affine motion estimates,
and dynamically makes the decision whether the motion is different enough
from pure translational motion for all the affine parameters to be
transmitted. I will attempt to use Lucas@Kanade algorithm for feature
tracking in order to obtain the motion estimates.

As a final result, I expect to have a rate-distortion performance
comparison of both algorithms for varying block sizes and varying
motion-estimation accuracy, as well as a time efficiency comparison.