Objectives of the Network
This project aims to develop motion analysis methods for image communication
systems. A large number of applications in image communication can benefit
from the results of the research within this network, including digital TV,
videophone, videoconferencing, and also publishing, medicine, computer-aided
manufacturing, computer graphics, remote expert consultation, and
telerobotics.
The network is involved in four directions:
- advanced motion modelling, and analysis;
- representation of the relevant information;
- implementation of the processing;
- objective criteria for the quality of image communication.
In motion analysis researchers are investigating the following problems and
methods: moving areas detection, 2D displacement field estimation and
segmentation using markovian stochastic models, or mathematical morphology
methods, or spatiotemporal Fourier analysis, and/or multiresolution analysis.
Global motion, as well as stereoscopic motion, is also under study. Schemes of
image prediction, filtering and interpolation using motion compensation form
part of the research. The project is also interested in parallel
implementation of motion analysis algorithms.
Results and achievements
The major results up to now (December 1995) are the following:
- moving object location methods using a Markov random field model formulation are developed.
- relaxation algorithms to regularize a motion field
taking into account its discontinuities are obtained.
- motion-based segmentation using morphological approach is proposed.
- recovery of apparent acceleration of moving objects in a scene is obtained.
- a method of temporal linking a spatio-temporal segmentation is proposed for
arbitrary region shapes.
- the 3-D camera motion is estimated using the 2-D displacement field.
- image layering is obtained by jointly using motion and luminance information.
- a 3-D human motion analysis system with interactive user interface is developed.
- genetic algorithms and their application to stereoscopic field estimation are studied.
- a relaxation algorithm for joint motion and disparity estimation in stereoscopic sequences is
proposed.
- a method for motion and disparity estimation in the presence of illumination variations
is proposed.
- methods for interpolating intermediate views with multiocular sensors are proposed.
- a new region-based discrete wavelet transform for image coding is developed.
- a new overlapped projection transform is introduced and geometrical vector quantization
is used for coding moving images.
- object-oriented coding systems are studied with particular focus on determining regions
of importance.
- a post-processing scheme that improves the subjective quality of compressed images
is proposed based on perceptual quality criteria.
- parallel implementations of motion-based video compression algorithms have been done on the CRAY
T3D.
- a fast play back is implemented on digital VCR.