Browse Prior Art Database

Advanced Motion Estimation for Moving Picture Experts Group Encoders

IP.com Disclosure Number: IPCOM000117670D
Original Publication Date: 1996-Apr-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 41K

Publishing Venue

IBM

Related People

Linzer, EN: AUTHOR

Abstract

Moving Pictures Experts Group (MPEG) encoding is discretionary -- different encoders produce different bit streams with the same video input. Motion fields are typically smoothed out by penalizing large motion vectors. This improves quality for slow moving scenes, but can actually degrade fast moving scenes.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 100% of the total text.

Advanced Motion Estimation for Moving Picture Experts Group Encoders

      Moving Pictures Experts Group (MPEG) encoding is discretionary
-- different encoders produce different bit streams with the same
video input.  Motion fields are typically smoothed out by penalizing
large motion vectors.  This improves quality for slow moving scenes,
but can actually degrade fast moving scenes.

      Motion estimation for this invention is performed by choosing
the motion vector with the smallest modified error.  The modified
error is described as the norm of the difference between the current
and reference blocks plus a position error.  The position error is an
increasing function of the size of the motion vector and a decreasing
function of the average motion vector used in a previously encoded
picture.  This approach can be used for any level of search in a
hierarchical estimation algorithm.

      A block diagram of a motion estimator using this technique is
shown in the Figure.  The block matcher computes match errors between
the current macroblock and macroblocks in the reference frame.  The
position error calculator computes errors for each motion vector
based on the motion vector size and the average motion vector from a
previous frame.  The position errors are added to the match errors to
compute the modified errors.  A min detector selects the best vector
based on a minimum modified error criterion.