Contact - General information - News - Research - Software - Publications
Ferréol Soulez
Centre de Recherche Astrophysique de Lyon
UMR 5574 / Université Lyon 1
9 avenue Charles André
F-69561 Saint Genis Laval Cedex
France
(+33) 4 78 86 85 46

ferreol dot soulez at univ-lyon1 dot fr
github.com/FerreolS

General information

I'm assistant astronomer in HARISSA group (High Angular Resolution, Imaging science, and Stellar Surroundings Astrophysics) at the observatory of Lyon. My research interests are in inverses problems approaches for signal and image restoration and reconstruction. I focus mainly on possibly space-variant deblurring and autocalibration (i.e. estimation of instrumental parameters from the observation as in blind deconvolution) specifically for multivariate signals (such as hyperspectral data, video, temporal sequences of observations) in a wide range of applications: astronomical imaging, medical imaging, digital holography, astronomical optical interferometry, microscopy, non destructive testing...

News

  1. Lecture about inverse problems for IOGS/TSE/Master AIMA in St Etienne. Slides are here .
  2. Lecture about inverse problem and deconvolution at "Stage détection" organized by labex FOCUS and held at OHP. Lab source code is here .
  3. Presentation of proximity operators for phase retrieval at the Inverse Problems Methods, Application and Synergies (IPMAS) workshop in Santiago (Chile).
  4. Lecture on blind deconvolution microscopy INSERM Workshop 249 in Bordeaux.
  5. http://ateliers249-inserm.evenium-site.com/site/atelier-de-l-inserm249
  6. Inverse problems lecture during FOCUS school at Observatoire de Haute Provence.
  7. 3D deconvolution microscopy workshop at MIFOBIO with Daniel Sage (EPFL/BIG).
  8. Module on image formation in microscopy at MIFOBIO illustrated with Icy protocols. Slides and protocols.
  9. Inverse problems lecture during FOCUS school at Observatoire de Haute Provence.
  10. Permanent assistant astronomer at CRAL, Observatoire de Lyon.
  11. Postdoc with M. Unser at Biomedical Imaging Group at EPFL (Switzerland) working on space varying PSFs estimation for Euclid space telescope (with F. Courbin from EPFL/LASTRO) and lensfree phase imaging (with EPFL/LAPD).
  12. My "L2DP3D" deconvolution method has reached the first place of the "Second International Challenge on 3D Deconvolution Microscopy" held during ISBI 2014 in Beijing.
  13. The deconvolution method I proposed for the "3D Deconvolution Microscopy ISBI 2013 Challenge" has reached the first place.
  14. Postdoc at CRAL on image reconstruction algorithms in hyperspectral optical interferometry in astronomy (ANR POLCA, Processing of pOLychromatic interferometriC data for Astrophysics).
  15. Postdoc at Centre Commun de Quantimetrie, an imaging facility of Lyon 1 University, working on restoration methods for biology and astronomy (funding ANR MiTiV, Méthodes Inverses de Traitement en Imagerie du Vivant).
  16. Postdoc at INSA Lyon on diffraction peaks extraction algorithms in energy dispersive X ray diffraction (EDXRD) in the context of luggage security screening (ANR research project SPIDERS).
  17. Phd Thesis defence in Saint-Etienne University (France). It focused on inverses problems methods for multidimensional data reconstruction.

Research

Research projects



Context

Spectral average of the reconstructed image  from GRAVITY simulated data Observation of complex objects using existing interferometers as well as the development of future instruments of the VLTI (GRAVITY, MATISSE) highlight the needs of algorithms dedicated to multi-spectral interferometry. In the context of POLCA project, I propose with E. Thiébaut an novel framework for image reconstruction in multi-spectral interferometry.

Key idea(s)

Performing this 3D (x,y,\lambda) reconstruction globally, the new MiRA-3D image reconstruction algorithm for polychromatic interferometry has two main advantages: i) it effectively uses the chromatic dispersion to increase the (u,v) coverage without assuming a "gray" object, ii) it uses the a priori knowledge that the observed object is somewhat spectrally correlated. These priors are enforced by means of spatio-spectral regularization such as spectral total variation or grouped sparsity that I have already proposed in the context of integral field spectrography. MiRA-3D is based on an ADMM framework which split the global optimization problem in a succession of smaller and easier problems. Solving each of these small problems rely on expressing the corresponding so-called "proximal operator" for both the data fidelity terms (with respect to complex visibilities, power spectrum and phase closures) and the regularization terms (total variation, spatio-spectral grouped sparsity ...)

Results

This new framework is very flexible: changing the prior and/or the type of interferometric is done by changing the coresponding prox operator. On GRAVITY simulation, the error on stars position is below 0.1 mas and the error on recosntructed star spectra is below 1%.

Related publications:

"An image reconstruction framework for polychromatic interferometry" F. Soulez and E. Thiébaut in Improving the performances of current optical interferometers - International Colloquium at haute provence observatory, St Michel l'observatoire : France Sept. 2013. (slides)
 


Context

Monomode optical interferometers get rid of atmospheric effects by integrating observables which are insensitive to the differential pistons between the interfering telescopes. For instance, the Fourier phase of the object is preserved in the so-called phase closures. This is however to the detriment of the amount of measured information.

Key idea(s)

Exploiting the chromatic behavior of the differential pistons, it is in principle possible to perform "a posteriori fringe tracking" to coherently add the complex visibilities of the short exposure fringe patterns and thus not only enhance the signal to noise ratio but also minimize the loss of information. To benefit from these improvements, there are several important difficulties. First, a global optimization has to be solved to estimate the instantaneous differential piston parameters from the available data. Second, specific image restoration methods have to be designed to deal with the resulting integrated observables.

Related publications:

"A posteriori finge sensing" F. Soulez, E. Thiébaut, M. Tallon and I. Tallon-Bosc in Improving the performances of current optical interferometers - International Colloquium at haute provence observatory, St Michel l'observatoire : France Sept. 2013. (slides)
 

 


Context

A good estimate of the PSF is mandatory to perform non blind deconvolution. In wide field fluorescence microscopy, such an estimate is derived of theoretical models or calibrated using fluorescent beads. However, using only known parameters about the setup (wavelength, NA, refractive index,...), theoretical models lack flexibility to represents the wide variety of PSFs. Similarly, measured PSFs are often noisy due to the small size of calibration beads, especially at high numerical aperture.

Key idea(s)

Blind deconvolution algorithms bypass the problem of PSF calibration by simultaneously estimating the PSF and the object. The 3D PSF is modeled after a parametrized pupil function. The PSF parameters are estimated jointly with the object in a maximum a posteriori framework.

Results

In wide field fluorescence microscopy, our method outperforms state-of-the art blind and non-blind deconvolution methods and gives a depth resolution close to the lateral resolution.

Related publications:

"Blind deconvolution of 3d data in wide field fluorescence microscopy" F. Soulez, L. Denis, Y. Tourneur and E. Thiébaut in 9th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Barcelona, Spain, May 2012. (pdf,slides)
 


2010 - present


Context

deconvolution of Hubble Space Telescope simulations Image deblurring is essential to high resolution imaging and is therefore widely used in astronomy, microscopy or computational photography. While shift-invariant blur is modeled by convolution and leads to fast FFT-based algorithms, shift-variant blurring requires models both accurate and fast. When the point spread function (PSF) varies smoothly across the field, these two opposite objectives can be reached by interpolating from a grid of PSF samples.

Key idea(s)

We developped a physically grounded model for PSF modeling that leads to a fast and accurate shift-variant blurring operator.

Results

We applied our model to simulations of Hubble Space Telescope data and shown good reconstruction performance.

Related publications:

"Fast model of space-variant blurring and its application to deconvolution in astronomy," L. Denis, E. Thiébaut, and F. Soulez, IEEE International Conference on Image Processing (ICIP), Brussels, September 2011. (pdf,poster)
 


Context

spatio-spectral cut of a restored integral field spectrograph data Integral field spectrographs are popular astronomical instruments that provide hyperspectral images. These data cube give much more information than monochromatic images but have a low signal to noise ratio.

Key idea(s)

Using a fast model of spectrally varying blur, we design a iterative restoration method taking into account specificity of real data (narrow field, high dynamic, non-stationary noise,..).

Results

Results show a significant gain both in term of resolution and in term of denoising with some field of view extrapolation.

Related publications:

“Restoration of hyperspectral astronomical data with spectrally varying blur,” F. Soulez, E. Thiébaut and L. Denis, to appear in EAS Publication series.
 
“3-D deconvolution of hyper-spectral astronomical data,” S. Bongard, F. Soulez, E. Thiébaut, and E. Pécontal Monthly Notices of the Royal Astronomical Society, 2011, 418 (1), pp 258-270.
 


Context

host galaxy extracted from the  observation sequence The Nearby Supernova factory use Type Ia supernovae as cosmogical probes to measure the expansion history of the Universe. To that aim, it is mandatory to extract precisely their spectrum and to follow its evolution using a sequence of integral field spectrographic observations. Unfortunately, flux from the supernova, from its host galaxy and from the sky background are mixed by the atmospheric turbulence.

Key idea(s)

We have defined different priors for each component of the data (SN, galaxy and sky): the SN is spectrally and temporally varying punctual source, the galaxy is spatially and spatially structured and does not vary with time, the sky background is spatially flat, the PSF varies spectrally and temporally according to physical laws,... Using these priors, each component is deconvolved and extracted from the (x,y,lambda,t) data,

Results

The host galaxy is restored even outside of the field of view of the IFU. The supernova spectra are extracted without bias leading to a better fit of the luminosity curves. This method doubled the number of unbiased supernovae that could be included in subsequent studies.

Related publications:

“4D deconvolution and demixing for supernova follow-up”, S. Bongard, E. Thiébaut and F. Soulez, proc. of WHISPERS 2009, Grenoble, August 2009.
 
“Multi-Wavelength, Multi-Epoch Integral Field Spectrograph Extraction for the Nearby Supernova Factory”, S. Bongard et al., 213th Meeting of the American Astronomical Society with HAD and HEAD, Long Beach, 2009.
 


Context

The study of dynamical phenomena involve the acquisition require acquisition of observation sequences with a very short exposure time. This results in low flux observations and therefore low signal to noise ratio. In addition, the imaging system inevitably introduces a blur that degrades the instrumental resolution observations.

Key idea(s)

The whole sequence being observed with the same instrument, one can assume that the blur varies slowly over time. With this assumption, we propose to use the entire sequence to constrain the PSF estimation.

Results

This blind deconvolution method had proven its capability and robustness in three types of modality: coronarography, confocal microscopy and bright field microscopy.

Related publications:

"Heterogeneous Multidimensional Data Deblurring", F. Soulez, E. Thiébaut, A. Gressard, R. Dauphin and S. Bongard, EUSIPCO, Lausanne, August, 2008. (pdf)
 
"Blind deconvolution of video sequences", F. Soulez, E. Thiébaut, Y. Tourneur, A. Gressard and R. Dauphin, IEEE International Conference on Image Processing (ICIP), San Diego, October, 2008. (pdf,slides)
 

 


Context

particle detection algorithm Digital holography is the method of choice for time-resolved 3D measurement of the location of particles in a flow. These measurements are crucial to validate numerical simulations of turbulence. The 3D location of several particles can be recovered from a single hologram by analyzing their diffraction patterns. Classically, this is performed in two steps: first, a 3D volume is reconstructed by simulating optical diffraction of the hologram. Then, the maximum of focus location of the image of each particle is detected. These approaches suffer from severe bias close to the hologram boundaries, and false detections occur due to multiple focusing or speckle noise.

Key idea(s)

Such drawbacks can be circumvent by following an inverse problem approach. Instead of reconstructing a 3D volume image from the hologram, particles are directly detected by matching the diffraction patterns on the hologram. An approach similar to the matching pursuit algorithm is proposed, with sub-pixel refinement by local optimization.

Results

The accuracy of the detection is improved by a factor 5 compared to that of classical techniques in a standard experimental configuration. Out-of-field detection is demonstrated, even far from the hologram boundaries.

Related publications:

"Inverse problem approach for particle digital holography: out-of-field particle detection made possible," F. Soulez, L. Denis, E. Thiébaut, C. Fournier, and C. Goepfert, J. Opt. Soc. Am. A, 24 (12), 3708-3716, 2007. (pdf, doi)
 
"Inverse problem approach for particle digital holography: accurate location based on local optimisation," F. Soulez, L. Denis, C. Fournier, E. Thiébaut, and C. Goepfert, J. Opt. Soc. Am. A, 24 (4), 1164-1171, 2007. (pdf, doi)
 
"Digital holography of particles: experimental parameters setting and benefits of the inverse problems approach," J. Gire, L. Denis, C. Fournier, C. Ducottet, E. Thiebaut, and F. Soulez, Meas. Sci. Tech., 19, 2008. (pdf, doi)
 

Software

Most of my Software are available on my github page, some of them are embeded in docker containers to be easily deployed on servers.

Optical long baseline interferometry

I'm PI of the model fitting and image reconstruction (MFIR) working group of the JMMC, the French Center for Infrared and Optical Interferometry. We are in charge of the development of the model fitting software LITpro and the GUI frontend for image reconstruction algorithms OImaging.

Deconvolution microscopy

Inverse problems libraries

I'm involved in the developement of two librairies for inverse problems:

Misc

Publications

This page is clearly inspired by the one of Loïc Denis. I thank him for his css stylesheet.