Nowadays, most post-processing techniques are based on differential imaging (DI). It consists in estimating the residual starlight (ideally without any circumstellar signal) and subtracting this estimate from the scientific image. This operation is repeated for every available image of the target star. The subtracted images are then combined to increase the signal-to-noise ratio of the potential circumstellar signals (planets and/or disks). A so-called detection map is then built (usually a signal-to-noise ratio map with a threshold), providing a confidence level for any detection in the field of view.
In order to estimate the starlight residuals, diversity is required: the circumstellar signals of interest must have different behavior or properties from those of the residual starlight noise, so that they can be disentangled. This diversity is usually achieved through specific observing strategies (e.g. polarization imaging, pupil tracking imaging, reference imaging etc.).
The image below summarizes the four steps of differential imaging towards the detection of exoplanet (or disk) signals:
High-contrast imaging general reviews
- Direct Imaging and Spectroscopy of Extrasolar Planets: Currie et al., 2023
- An Introduction to High Contrast Differential Imaging of Exoplanets and Disks: Follette et al., 2023
- Peering through SPHERE Images: A Glance at Contrast Limitations: Cantalloube et al., 2019
Post-processing techniques dedicated to high-contrast imaging
The paper presented below correspond to data processing based on Angular Differential Imaging (ADI). The concept of ADI applied to high-contrast imaging has been introduced by Marois et al., 2006.
In order to use ADI-based techniques, the images are taken in pupil tracking mode during the observation sequence (usually about an hour of observation, with exposure times ranging from milliseconds to minutes). This observing strategy introduces angular diversity: during the whole observation sequence (1) the astrophysical scene rotates around the optical axis (centred on the target star) following the parallactic angle, while (2) the telescope pupil (entrance aperture) remains fixed, meaning that the optical aberrations are kept in the same direction in the field of view during the observation sequence.
The image below summarizes how a typical so-called ADI data set is acquired:
1/ Classical Speckle Subtraction techniques
- LOCI: Lafrenière et al., 2007
- KLIP (PCA): Soummer et al., 2012
- SVD (PCA): Amara & Quanz, 2012
- NMF: Ren et al., 2018
2/ Advanced Subtraction techniques
- RSM: Dahlqvist et al., 2020
- STIM-map: Pairet et al., 2019
- LR-map: Daglayan et al., 2022
- LRPT: Vary et al., 2023
- AMAT: Daglayan et al., 2024
3/ Inverse Problem approaches
- ANDROMEDA: Cantalloube et al., 2015 (after Mugnier et al., 2009)
- FMMF: Ruffio et al., 2017 (after Pueyo et al., 2016)
- PACO: Flasseur et al., 2018
- TRAP: Samland et al., 2021
- SNAP: Thompson et al., 2021
4/ Supervised Machine learning
- SODDIN: Gomez-Gonzalez et al., 2017
- NA-SODDIN: Cantero et al., 2023
- HSR: Gebhard et al., 2022
- Deep-PACO: Flasseur et al., 2023
5/ Dedicated to extended structure imaging
- NMF: Ren et al., 2018
- DISK-FM: Mazoyer et al., 2020
- MAYONNAISE: Pairet et al., 2021
- REXPACO: Flasseur et al., 2021
- IADI: Stapper & Ginksi, 2022
- MUSTARD: Juillard et al., 2023
Detection limits & performance assessement
- Confidence level and Sensitivity limits in HCI: Marois et al., 2007
- Fundamental limitations of HCI set by small sample statistics: Mawet et al., 2014
- A New Standard for assessing the performance of HCI: Jensen-Clem et al., 2017
- Robust Detection Limits for HCI in the presence of non-Gaussian noise: Bonse et al., 2023