RESEARCH INTERESTS
-
Deep learning for image and video compression
-
Compressed sensing optimization method and its application to video
watermarking.
- Functional
analysis, image decomposition models (Structure/texture), analysis/synthesis
and inpainting in a hierarchical approach for image and video
analysis and compression.
- Regularization schemes for image and video restoration,
compression, denoising and analysis.
- Wavelets for image and motion analysis (redundant,
multiresolution, 2D+T, 3D+T): Cauchy, conical and
spherical wavelets
and their developments (collaboration with UCL Université
Catholique de Louvain-La-Neuve).
- Segmentation of 2D+T video sequences by a Bayesian approach and
Hidden Markov Model (HMM) of Potts-Markov type
(see thesis below).
-
Motion
Quantification based on segmentation and region labeling.
Application to the analysis of biocellular motions (see thesis
below).
- Motion estimation and video
compression with motion-tuned spatio-temporal continuous
wavelets (see thesis below).
- Fast trajectory identification for scene analysis and video
compression (see thesis below).
- Shape
recognition by wavelets multiresolution analysis and curvature
comparison.
- Analysis of STM (Scanning Tunneling Microscope) images with
wavelets and ridgelets multiresolution analysis.
- Motion control of a mobile robot by application of flatness
property. Trajectory generation in presence of obstacles by the
potentials method (harmonic functions).
- Fractal and multifractal analysis applied to the diffusion
function of random propagation channels in the ionosphere
(Engineer thesis LETTI INT/Upsud).
Measurement of the monofractal state by the
Bouligand-Minkowski dimension (BCM method) and previsional
determination of the ionosphere state for transmission.
Wavelet Transform Maxima Method (WTMM /
MMTO) and multifractal spectrum determination for 3D ionosphere
characterization.
- Theory of quantum states in solids and semi-conductors. Quantum
electronics :
Theoretical model and automatic computation of band and miniband
structure in GaAlAs-GaAs superlattices, in a non-parabolic
approximation and based on Kane's eight bands model (CNET Bagneux
1987).
Spectral-conductivity analysis of the perpendicular propagation in
the base of a TBS (Superlattice Bipolar Transistor).
(Microelectronics DEA training period; CNET Bagneux, 1987).
Thesis
(Doctoral school :
STITS, Sciences et Technologies de l'Information des
Télécommunications et des Systèmes, Université XI Paris-Sud,
UFR d'Orsay.)
Title : "Image Segmentation and Motion
estimation"
Part I : Motion
Estimation with spatio-temporal wavelets tuned to motion.
Part II : Segmentation and Motion Estimation using HMM (Hidden
Markov Models) and a Bayesian approach in the direct
(pixels/voxels) and wavelet domains.
Part
I (IEF, Orsay Paris-Sud)
Introduction of a new,
contextual, scheme for compression and analysis of video scenes
based on :
- A family of spatio-temporal wavelets tuned to motion (MTSTWT :
Motion-Tuned Spatio-Temporal Wavelets)
- The fast identification of objects trajectories based on objects
motion parameters quantized by the MTSTWT.
- A motion prediction based on identified trajectories.
http://djafari.free.fr
Resolution of inverse problems, and in particular
segmentation, by using a Bayesian approach and Hidden Markov
Models (PMRF, Potts-Markov Random Field) applied to images and
sequences in a direct and a transformed domain (wavelets
domain) :
- Fast,
iterative, Bayesian segmentation applied to 2D+T sequences.
- Motion estimation in video
sequences, based on the former 2D+T Bayesian segmentation.
- Speed reduction of the
Bayesian segmentation in the direct domain by performing the
segmentation in the wavelet coefficients domain, with the PMRF
model enhanced to fit to the priviledged orientations of the
wavelet subbands.
The meaning of
probability.
