| | |
Summary: M.I.T Media Laboratory Perceptual Computing Section Technical Report No. 344
To appear: 1995 IEEE Workshop on Neural Networks for Signal Processing
Motion Estimation and Segmentation Using a
Recurrent Mixture of Experts Architecture
Yair Weiss and Edward. H. Adelson
Dept. of Brain and Cognitive Sciences and the Media Lab
MIT E15-384, Cambridge, MA 02139, USA
fyweiss,adelsong@media.mit.edu
Abstract
Estimating motion in scenes containing multiple
motions remains a dicult problem for computer
vision. Here we describe a novel recurrent net-
work architecture which solves this problem by
simultaneously estimating motion and segment-
ing the scene. The network is comprised of lo-
cally connected units which carry out simple cal-
culations in parallel. We present simulation re-
sults illustrating the successful motion estima-
tion and rapid convergence of the network on real
image sequences.
|
