Share Email Print

Proceedings Paper

Motion-estimation parallel algorithm based on band matching
Author(s): Sausan Yazji; Magdy A. Bayoumi; Bertrand Zavidovique
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

A new algorithm to compute the motion vector has been developed based on the block-matching approach in which bands are used instead of blocks. The new band-matching algorithm (BDMA) offers better performance than both the fixed size and variable-size block-matching algorithms because it uses less vectors and processes parts of the image instead of the whole image. The BDMA extracts the likely moving objects, forms the image of differences between initial successive frames in an image sequence, divides each 'object' into bands with different sizes, forms a 4 by 4 block around each edge in the band, looks for the type of edges, calculates the different of the textures inside the band, and starts the matching process between frames. Bands are matched according to the number of edges and their type (the type is created and stored using some kind of code-book), and the texture values. Matching results produce a specific vector for the estimated motion. To enhance the performance two independent external memory are employed, they are accessed in parallel. One memory is indexed by the other, one to store the ongoing images, and the second to buffer the bands and results. There sizes and features are adapted to support the application for which the circuit is needed. The new algorithm has been evaluated using several standard video images.

Paper Details

Date Published: 19 September 1997
PDF: 13 pages
Proc. SPIE 3166, Parallel and Distributed Methods for Image Processing, (19 September 1997); doi: 10.1117/12.285556
Show Author Affiliations
Sausan Yazji, Univ. of Southwestern Louisiana (United States)
Magdy A. Bayoumi, Univ. of Southwestern Louisiana (United States)
Bertrand Zavidovique, Univ. of Southwestern Louisiana (France)

Published in SPIE Proceedings Vol. 3166:
Parallel and Distributed Methods for Image Processing
Hongchi Shi; Patrick C. Coffield, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?