Data Assimilation for Image Processing
Ill-Posed problems in Image Processing
The Ill-posed problems are generally solved using regularization technics such as the Tikhonov method which constrains spatially the solution. We show how data assimilation can be used as an alternative of Tikhonov method to solve ill-posed problems. Data Assimilation owns natural mechanisms to integrate a temporal model of evolution of structures within images and to manage missing data.
People involved
Isabelle HerlinRelated publications
RR6477.Optical flow estimation
3D reconstruction
People involved
Carlos Dantas De AlmeraRelated publications
Bio-Medical Image Analysis
In September 2001, I started researchs on Computer Vision applied to Biomedical images as associated professor at LIP6.
Reconstruction 3D+t
People involved
Wafa Rekik (core of her PhD thesis), Séverine Dubuisson.Related publications
CAIP07Image analysis on spherical structures in images
Reconstruction 2D in multimodal data and missing data
Given several images acquired from various fluorescence modes at several times, we are interested to reconstruct an exhaustive image sequence. Variations around the particular filtering have been proposed.
Involved people
Abir El Abed (core of her PhD thesis), Séverine Dubuisson (Abir's director).Related publications
ACVIS07, ICIP07, GRETSI07, VISAPP07, ACVIS06.Remote Sensing Imaging
Processing Ikonos image for coconuts management
Involved people
Raimana Teina (core of his PhD thesis), Benoit Stoll (co-director).Related publications
MAJECSTIC07, IGARSS08.Segmentation
Markov random field technics can be used to segment homogenous regions inside images. We have proposed a markovian model using some grey level values homogeneity properties within region to segment. In addition, spatial gradient and temporal information can be used in order to enhance the quality results too.
A first model was initially developped for segmention of cardiac cavity inside echocardiographic image sequences. The model integrates spatio-temporal properties in order to track the cavity deformation along the sequence.
Then the model has been adapted for segmenting cloud structures inside meteorological data (typically, data acquired by METEOSAT). A spatial model and a temporal model had been developped.
Finally, the model was adapted for Interferometric images to segment homogenous regions.
Related publications
ECCV94, RR94, ICPR98, ICPR00.Motion Estimation
We are interested in computation of the displacement of cloudy structure inside meteorological data acquired by METEOSAT. Most of motion estimation method are based on point brightness invariance principle. A such hypothesis is not necessary verified on INFRARED data. Cloud elevation are related to their temperature. When a change of elevation occurs, the brightness invariance is no more true on infrared data. We propose a new approach taking in account physical properties of infrared data and using the mass conservation principle verified by perfect fluids as clouds. It leads to a new motion constraint that we have called the ``total brightness invariance''.
Actually, this constraint is known as the Extended Optical Flow Constraint already studied by several authors in the past. We give to this equation a new image signification and, in the meteorological context, a new physical signification.
Related publications
APMS98, ICASSP99, CVPR00, ICPR00, EMS00.Trajectory Analysis
We propose a method to caracterize velocity vector field. We propose to project the velocity vector field or the pixels trajectory (integration of the velocity vector field) on Fourier base. Then, some data analysis technics such as PCA are used to analyze trajectories (see Trajectory analysis).