PaperlayerView exact revision on arXiv

Source-linked AI summary

VIP: Vortex Image Processing package for high-contrast direct imaging

C. A. Gomez Gonzalez, O. Wertz, O. Absil, V. Christiaens, D. Defrere, D. Mawet, J. Milli, P. -A. Absil, M. Van Droogenbroeck, F. Cantalloube, P. M. Hinz, A. J. Skemer, M. Karlsson, J. Surdej

arXiv:1705.06184v1astro-ph.IM

TL;DR

High-contrast exoplanet imaging needs flexible tools for processing challenging ADI data and separating companion signals from stellar speckles. VIP addresses this need with a modular Python library spanning preprocessing, PSF subtraction, source estimation, and sensitivity analysis; applied to HR8799, it found no significant additional point source beyond the four known planets.

  • Problem

    Direct exoplanet imaging is challenging, and ADI processing requires methods for constructing reference PSFs and extracting faint companions from stellar speckles.

  • Method

    VIP provides an instrument-agnostic Python framework combining preprocessing, ADI post-processing, companion position and flux estimation, and sensitivity-curve generation.

  • Results

    Applied to a deep HR8799 sequence, VIP found no significant additional point-like source beyond the four known planets.

  • Takeaways & Limitations

    VIP is presented as an open-source library for assembling high-contrast imaging workflows and processing large ADI datasets with limited memory.

  • Takeaways & Limitations

    VIP provides contrast curves rather than ROC-based performance analysis, whose proper computation and use remain ongoing research.

Abstract

from arXiv · show

We present the Vortex Image Processing (VIP) library, a python package dedicated to astronomical high-contrast imaging. Our package relies on the extensive python stack of scientific libraries and aims to provide a flexible framework for high-contrast data and image processing. In this paper, we describe the capabilities of VIP related to processing image sequences acquired using the angular differential imaging (ADI) observing technique. VIP implements functionalities for building high-contrast data processing pipelines, encompass- ing pre- and post-processing algorithms, potential sources position and flux estimation, and sensitivity curves generation. Among the reference point-spread function subtraction techniques for ADI post-processing, VIP includes several flavors of principal component analysis (PCA) based algorithms, such as annular PCA and incremental PCA algorithm capable of processing big datacubes (of several gigabytes) on a computer with limited memory. Also, we present a novel ADI algorithm based on non-negative matrix factorization (NMF), which comes from the same family of low-rank matrix approximations as PCA and provides fairly similar results. We showcase the ADI capabilities of the VIP library using a deep sequence on HR8799 taken with the LBTI/LMIRCam and its recently commissioned L-band vortex coronagraph. Using VIP we investigated the presence of additional companions around HR8799 and did not find any significant additional point source beyond the four known planets. VIP is available at http://github.com/vortex-exoplanet/VIP and is accompanied with Jupyter notebook tutorials illustrating the main functionalities of the library.

1. INTRODUCTION

Directly imaging exoplanets is challenging but complements indirect detection, and ADI is a widely used strategy supported by multiple PSF-subtraction approaches. VIP packages these capabilities into an instrument-agnostic, extensible Python library for high-contrast imaging.

  • 1. INTRODUCTION: Direct exoplanet imaging remains challenging, with only a few tens of exoplanets directly resolved despite more than three thousand confirmed discoveries.Technological advances in near-infrared instruments, adaptive optics, and coronagraphy have enabled recent progress.
  • 1. INTRODUCTION: ADI is the most commonly used high-contrast imaging strategy, despite limitations, and is the focus of this paper.VIP supports processing workflows centered on ADI observations.
  • 1. INTRODUCTION: ADI post-processing methods include LOCI, ANDROMEDA, PCA-based algorithms, and low-rank, sparse, and noise decompositions that rely on model PSFs.These approaches construct reference PSFs to improve detectability of planets and disks against stellar speckles.
  • 1. INTRODUCTION: VIP provides a flexible, instrument-agnostic Python toolbox with pre- and post-processing algorithms for angular, reference-star, and multispectral differential imaging.The library is modular, easy to use, documented, and released as open source.
  • 1. INTRODUCTION: The paper covers VIP’s design, preprocessing, ADI reference-PSF subtraction, source extraction, sensitivity estimation, and an on-sky demonstration.The sections collectively describe the library’s processing and evaluation workflow.

2. PACKAGE OVERVIEW

VIP is organized as a modular Python library built on the scientific Python ecosystem rather than as a fixed pipeline. Its subpackages support data handling, preprocessing, photometry, sensitivity estimation, and ADI post-processing.

  • 2. PACKAGE OVERVIEW: The library relies on open-source scientific packages including NumPy, SciPy, Matplotlib, Astropy, scikit-learn, pandas, and scikit-image.These dependencies support VIP’s scientific software implementation.
  • 2. PACKAGE OVERVIEW: VIP is a modular library whose users select procedures and their order instead of following a predefined linear workflow.Its organization supports extension and reuse through subpackages, modules, functions, and classes.
  • 2. PACKAGE OVERVIEW: VIP’s fits subpackage handles FITS files and enables loading high-contrast imaging datacubes from disk.It also supports displaying NumPy arrays through an interface to SAOImage DS9.
  • 2. PACKAGE OVERVIEW: The photometry subpackage provides signal-to-noise estimation, source detection, fake-companion injection, and sensitivity-limit computation.Planet signal-to-noise uses a small-sample-statistics approach based on a two-sample t-test over resolution elements.
  • 2. PACKAGE OVERVIEW: The preproc, llsg, madi, pca, and negfc subpackages contain low-level preprocessing and ADI post-processing algorithms.These subpackages organize the processing functions described for ADI datasets.

3. PRE-PROCESSING

VIP provides preprocessing operations for calibrated high-contrast imaging datacubes, including registration, resizing, shifting, cropping, temporal combination, and bad-pixel or bad-frame handling. The library requires users to perform basic calibration such as flat-fielding and dark subtraction externally.

  • 3. PRE-PROCESSING: VIP requires datacubes that have already undergone basic calibration, including successful flat-fielding, while leaving heterogeneous instrument-specific calibration to users.This preserves the library’s instrument-agnostic design.
  • 3. PRE-PROCESSING: Preprocessing supports resizing, upscaling, pixel binning, shifting, rotating, cropping, temporal subsampling, and sequence collapse by mean, median, or trimmed mean.The functions operate on individual images and image sequences.
  • 3. PRE-PROCESSING: Frame registration is critical for ADI because the star must remain centered and all frames must be well aligned.VIP supports registration using several image-based procedures, including Gaussian or Moffat fitting and Fourier cross-correlation.
  • 3. PRE-PROCESSING: VIP includes procedures for detecting bad pixels and bad frames, with bad pixels replaced using neighboring-pixel medians.These operations complement geometric registration and other low-level preprocessing steps.

4. POST-PROCESSING

VIP provides multiple ADI reference-PSF subtraction methods, including classical median subtraction, PCA variants, NMF, and LLSG, with parameter optimization and extensions for diverse data formats.

  • 4.1. Median reference PSF subtraction: Classical ADI builds a median reference PSF, while annular processing selects temporally nearby frames after enforcing a separation-dependent parallactic-angle threshold.The threshold is defined using the PSF FWHM, a user-set δ, and angular separation r; median subtraction has limited small-angle performance.
  • 4.2. PCA-based algorithms for reference PSF subtraction: PCA-based processing can also be applied to RDI, multispectral SDI, and IFS data, although VIP’s RDI and SDI processing remained ongoing work in this paper.These extensions are noted but not described in detail.
  • 4.2. PCA-based algorithms for reference PSF subtraction: VIP optimizes the PCA component count by grid-searching the S/N at a specified image location, with adaptive refinement to avoid unpromising parameter regions.For HR8799e, the mean S/N in a FWHM aperture was maximized with 16 PCs, although maximum S/N did not coincide with maximum throughput.
  • 4.2. PCA-based algorithms for reference PSF subtraction: VIP’s PCA implementations include full-frame, annular, and library-selection variants that reduce companion self-subtraction by excluding insufficiently rotated frames.Library refinement improves signal preservation but can require n singular value decompositions, increasing computational cost.
  • 4.3. Non-negative matrix factorization for ADI: NMF models ADI data with non-negative low-rank factors and produces final images very similar to full-frame ADI-PCA, providing a complementary robustness test for detections.NMF components are strictly positive, but the non-convex optimization is computationally harder than PCA and lacks a guaranteed optimum.
  • 4.4. LLSG for ADI: LLSG achieves higher point-source S/N and better ROC-space performance than full-frame ADI-PCA, especially at small inner working angles, without greatly increasing computational cost.Its detectability advantage is attributed to complex speckle noise that limits full-frame ADI-PCA near the star.

5. FLUX AND POSITION ESTIMATION FOR ADI

VIP estimates companion position and flux with negative fake companion injection, iterative cancellation, and sequential refinement of the relevant parameters.

  • 5. FLUX AND POSITION ESTIMATION FOR ADI: NEGFC estimates companion position and flux by injecting a negative PSF template into calibrated frames and minimizing residuals after post-processing.Because injection occurs before processing, the method accounts for photometric and astrometric biases introduced by the reduction algorithms.
  • 5. FLUX AND POSITION ESTIMATION FOR ADI: The NEGFC objective is evaluated within a circular aperture centered on the companion, with a default radius of 4× FWHM and alternatives for speckle-rich regions.VIP can minimize pixel standard deviation instead of sum, or evaluate residual apertures frame by frame rather than after collapse.
  • 5. FLUX AND POSITION ESTIMATION FOR ADI: VIP first estimates flux on a grid with position fixed, then jointly refines radius, angle, and flux using downhill simplex minimization.The simplex refinement improves parameter determination but does not provide error bars.

6. SENSITIVITY LIMITS

VIP estimates sensitivity limits as detectable planet/star contrast versus separation, using small-sample corrections and empirically measured algorithm throughput. The paper cautions that contrast curves depend on post-processing and tuning, while ROC-based evaluation may better compare algorithms.

  • Detectable contrast is reported as a function of separation from the star, with a student-t correction accounting for small-sample statistics.
  • VIP builds contrast curves by measuring residual noise, injecting fake companions, and correcting for the chosen algorithm’s signal attenuation.Throughput is estimated as recovered fake-companion flux divided by initially injected flux.
  • Contrast curves vary with the post-processing algorithm and its tuning, and VIP therefore leaves ROC-based performance comparison to ongoing research.The paper identifies ROC figures of merit as better suited for comparing post-processing algorithms.

7. APPLYING VIP TO ON-SKY DATA

VIP was applied to HR8799 data through calibration, frame-quality selection, and multiple ADI post-processing methods. The analysis identified an instrumental ghost, found no significant fifth companion, and showed that more complex PSF subtraction improved inner-image cleaning while sensitivity depended on algorithm and parameters.

  • Data processing: The apparent companion near HR8799e was attributed to a secondary-reflection artifact, so the first batch of lower-quality frames was discarded.The artifact was bright in the first half, absent in the second, and located at the measured reflection offset.
  • Results: No significant point source beyond HR8799’s four known planets was detected after processing the datacube with several tuned ADI algorithms.The sequence was temporally subsampled for the showcase, producing a 499-frame datacube without pixel binning.
  • Sensitivity limits and discussion: More complex PSF subtraction techniques outperformed classical median subtraction for cleaning the innermost region near 2λ/D.The authors declined broader algorithm comparisons because robust benchmarking would require diverse datasets and community-defined metrics.
  • Sensitivity limits and discussion: Annular ADI-PCA achieved impressive contrast below 0.5 arcsec and had less dependence on the number of principal components than full-frame ADI-PCA.The cited sensitivity analysis also compares available ADI algorithms using 5-sigma contrast curves.
  • Sensitivity limits and discussion: Within 1 arcsec, the L-band vortex coronagraph improved contrast by up to 1 magnitude relative to the comparison dataset without the coronagraph.At small separations, the VIP full-frame ADI-PCA curve was more pessimistic, partly attributed to the applied student-t correction.

8. SUMMARY

VIP provides instrument-agnostic high-contrast imaging data processing, including low-rank ADI methods and an incremental PCA algorithm for larger-than-memory datasets. Applied to HR8799, VIP found no significant additional point-like sources beyond the four known companions.

  • 8. SUMMARY: VIP was tested on data from Keck/NIRC2, VLT/NACO, VLT/VISIR, VLT/SPHERE, and LBT/LMIRCam as an instrument-agnostic library.
  • 8. SUMMARY: VIP processes high-contrast imaging data from pre-processing through contrast-curve calculation, including PCA, NMF, and LLSG-based ADI post-processing.Its incremental ADI-PCA implementation targets big, larger-than-memory datasets.
  • 8. SUMMARY: Figure 7 reports 5-sigma sensitivity curves for full-frame ADI-PCA across PC counts and for selected VIP ADI algorithms after removing the four known companions.The curves include the small sample statistics correction.
  • 8. SUMMARY: VIP processing of a long HR8799 LBTI/LMIRCam sequence found no significant additional point-like sources beyond the four known companions.

A. DETAILS ON THE FULL-FRAME ADI-PCA ALGORITHM

Full-frame ADI-PCA represents the image sequence as a frame-by-pixel matrix, models the reference PSF with selected principal components, subtracts it, and derotates and combines the residuals. VIP implements equivalent eigen-decomposition and SVD routes, including memory-saving formulations.

  • A. DETAILS ON THE FULL-FRAME ADI-PCA ALGORITHM: Full-frame ADI-PCA loads frames into a matrix, selects k principal components, subtracts the resulting low-rank PSF model, then derotates and median-combines residuals.Mean-centering or standardization can precede component selection.
  • A. DETAILS ON THE FULL-FRAME ADI-PCA ALGORITHM: The principal components can be obtained from the dominant eigenvectors of the covariance matrix or from the dominant right singular vectors of the frame matrix.
  • A. DETAILS ON THE FULL-FRAME ADI-PCA ALGORITHM: VIP accelerates PCA by decomposing the smaller matrix MM^T or by taking the SVD of M^T, both yielding the same result as the corresponding full computation.These alternatives reduce the cost of working with large matrices.

B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS

Large ADI datacubes can exceed available memory and make conventional PCA impractical, while temporal or spatial subsampling can smear signals. VIP therefore provides randomized and incremental PCA variants, and experiments favor using the full sequence for signal-to-noise.

  • B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS: Temporal or spatial subsampling reduces computation time but can smear the signal, motivating methods that avoid subsampling for large datasets.
  • B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS: Figure B2 compares memory usage and processing time for full-frame ADI-PCA variants on a 10 GB datacube.For short or subsampled sequences, LAPACK SVD can make processing-time differences negligible.
  • B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS: Using the whole approximately 10,000-frame sequence without temporal subsampling produced the best recovered-companion S/N in the described injection experiment.The test used 21 PCs and measured mean S/N in a λ/D aperture.
  • B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS: VIP adds randomized SVD and incremental PCA options to reduce computation time and memory use when subsampling is undesirable.Randomized SVD approximates the decomposition through random projections, while incremental PCA processes batches online.
  • B. FULL-FRAME ADI-PCA FOR BIG ADI DATASETS: Incremental PCA processes large ADI datasets in batches through memory mapping, updates the principal components sequentially, and reconstructs residuals in a second pass.This reduces memory consumption for datacubes larger than available RAM.

C. ANNULAR ADI-PCA

Annular ADI-PCA divides frames into annuli and performs localized PCA with separation-dependent frame selection. Its parallel implementation exploits smaller patches and can adapt the component count to local residual statistics.

  • C. ANNULAR ADI-PCA: Annular ADI-PCA builds a separate matrix for each annulus and applies PCA after removing frames that fail the annulus-specific rotation threshold.
  • C. ANNULAR ADI-PCA: The annular residuals are assembled into a residual datacube, derotated to a common north, and median-combined into the final image.
  • C. ANNULAR ADI-PCA: Annular processing supports multiprocessing across zones, while smaller patches account for differing pixel statistics across the image.
  • C. ANNULAR ADI-PCA: The algorithm can automatically choose the number of principal components for each patch by minimizing the residual standard deviation.
Loading 1705.06184v1…