Post-processing for AO-corrected images Inverse Problems in Imaging - - PowerPoint PPT Presentation

1

Post-processing for AO-corrected images

Inverse Problems in Imaging

May 23rd, 2018 Summer school - Lake Como

24/05/2018 24/05/2018

Romain JL. FETICK Supervisors Thierry FUSCO (ONERA/LAM) Laurent MUGNIER (ONERA) Co-supervisors Benoit NEICHEL (LAM) Aurélie BONNEFOIS (ONERA)

2

Summary

• I. Imaging with adaptive
• ptics in astronomy
• II. Theory on deconvolution

and myopic deconvolution

• III. PSF parametrization
• IV. Tests on VLT SPHERE

and MUSE data

• V. Future developments

Paris Marseille Como Toulouse

3

Observation with Adaptive Optics

Object Blurred Noisy Image Atmosphere, Telescope AO system, CCD

CCD WFS

• Residual turbulence with AO still

produces “blurring“

• Photon and detector read-out noise
• Better images required (noiseless,

sharp edges, visible structures)

• Observation of satellites and asteroids
• Adaptive Optics (AO) partially corrects

the atmospheric turbulence

• AO greatly improves imaging

performances

Context Issues

Post-processing required to further improve image quality

4

Deconvolution

Estimated object Object Blurred Noisy Image Observed PSF

Deconvolution MISTRAL

Atmosphere, Telescope AO system, CCD Star

CCD WFS

Data fidelity A priori

→ Noise statistics → A priori on object

Criterion to minimize

• Observed PSF may significantly differ

from the real system PSF

• Errors on estimated object

5

Myopic deconvolution

Estimated object and PSF Object Blurred Noisy Image Observed PSF

Deconvolution MISTRAL

Atmosphere, Telescope AO system, CCD Star

CCD WFS

A priori

→ Noise statistics → A priori on object → PSD on the PSF

Criterion to minimize

Myopic term Data fidelity

• Better results than fixed-PSF
• High number of unknowns

6

Solution: parametrize the PSF

AO correction quality Turbulence Kolmogorov Strehl Diffraction & static artifacts Data Moffat MPHD 500ηm 600ηm 700ηm 800ηm 900ηm

VLT / MUSE VLT / SPHERE MPHD PSF modelling

7

PSF from star

Deconvolution MISTRAL PSF

t t+Δt

Star

→Deconvolutional issues: mismatch true PSF (linked to object) / observed PSF (star)

Asteroid

Image

Estimated object

State of the art

Image VLT - SPHERE Credit: P. Vernazza (LAM)

Protected Data Protected Data

8

Myopic deconvolution

Myopic Deconvolution MISTRAL PSF

t t+Δt

Star

→Deconvolutional issues →Low statistical contrast: data (N pixels) << unknown (2 x N)

Asteroid

Image

Estimated PSF Estimated object

State of the art

Statistics on the PSF variations

Protected Data Protected Data

9

Parametric myopic deconvolution

Parametric Myopic Deconvolution MISTRAL

t

Telemetry

→Suitable PSF for deconvolution → Physical meaning of the PSF’s parameters →High statistical contrast: data (N pixels) ≈ unknown (N+10) → No need for on-sky PSF observation: astronomical & military usage

Asteroid Estimated PSF

Parametric PSF

Goals of the thesis

Estimated object

Fried Strehl Jitter...

Statistics on the parameters variations

Protected Data

Image

Protected Data

10

Work objectives

• Evaluate improvements of

deconvolution using parametric PSF

• Stability toward parameters
• Myopic deconvolution
• Apply method on data

– Simulations – VLT (asteroids) – ONERA (satellites)

• Error maps on estimated
• bjects

Protected Data

11

Toulouse