Share Email Print
cover

Proceedings Paper

Compact motion representation based on global features for semantic image sequence coding
Author(s): Claude Labit; Henri Nicolas
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

For dynamic scene analysis, numerous studied developed derivation methods of qualitative motion features from a dense apparent velocity vector field (optical flow). However, this qualitative information cannot be used for image sequence reconstruction because no explicit reconstruction criterion is introduced within the feature identification process itself. This study proposes a new algorithmic approach which tries to solve the crucial trade-off between a small amount of motion parameters--viewed as assistance data in a motion compensated interpolation scheme for image sequence coding--and high quality of reconstruction. This approach could be described as a tentative step towards semantic coding scheme using global--and thus compact-- motion labels which characterize a restricted class of 3-D physical motions even if no 3-D measurements are required. This paper is structured as follows: after a brief introduction where the general framework is described, we present a hierarchy of affine models for motion field representation and the explicit derivation of global motion features associated with a simplified version of a linear model. The identification procedure, essentially based on minimization techniques using steepest descent methods, is shown. After all these initial requirements, we introduce the more efficient and new algorithmic step; it is concerned with the use of a deterministic relaxation scheme to smooth the motion vector field and improve the accuracy of the motion estimator, especially within regions where large magnitudes of motion are located. We also introduce a temporal linking operator which appears quite natural in the framework of iterative estimation methods where initial values are required. Finally, results on real image sequences are given; performances are expressed in terms of convergence speed, reconstruction quality, compactness of the motion representation and qualitative or quantitative efficiency of the motion estimation scheme.

Paper Details

Date Published: 1 November 1991
PDF: 12 pages
Proc. SPIE 1605, Visual Communications and Image Processing '91: Visual Communication, (1 November 1991); doi: 10.1117/12.50260
Show Author Affiliations
Claude Labit, IRISA/INRIA (France)
Henri Nicolas, IRISA/INRIA (France)


Published in SPIE Proceedings Vol. 1605:
Visual Communications and Image Processing '91: Visual Communication
Kou-Hu Tzou; Toshio Koga, Editor(s)

© SPIE. Terms of Use
Back to Top