Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
Parallel Predictive Motion Estimation using Object Recognition Methods
 

Summary: Parallel Predictive Motion Estimation using Object
Recognition Methods
Holger Blume, Aishy Amer
Universität Dortmund, Lehrstuhl für Nachrichtentechnik, Tel.: +49 231 755 4535
AG Schaltungen der Informationsverarbeitung Fax: +49 231 755 3196
44221 Dortmund
1.) Introduction
Motion estimation is one of the key techniques in modern digital video signal processing.
High quality vector fields are as well needed for vector based upconversion as for motion
compensated coding or noise reduction [7]. Most of the estimation algorithms which are used
today and which are implemented in dedicated hardware are block based algorithms (block
matching algorithms). Although they are easy to implement and the quality of their resulting
vector fields is better than that of the methods which were used formerly (gradient methods,
phase plane correlation) they still suffer from some drawbacks. Most important is that using
block based algorithms a block pattern will arise in the vector field (block pattern noise).
Real objects in real scenes do not coincide with block boundaries. Therefore object
recognition methods shall be introduced into motion estimation. Within this work an object
oriented motion estimation algorithm is presented which combines motion information and
object information and yields nevertheless a good computational performance.
2.) The new object based algorithm

  

Source: Amer, Aishy - Department of Electrical and Computer Engineering, Concordia University

 

Collections: Engineering