A parallel and automatically tuned algorithm for multispectral image deconvolution

Abstract

In the era of big data in the radio astronomical field, image reconstruction algorithms are challenged to estimate clean images given limited computing resources and time. This article is driven by the extensive need for large scale image reconstruction for the future Square Kilometre Array (SKA), the largest low-and intermediate frequency radio telescope of the next decades. This work proposes a scalable wideband deconvolution algorithm called MUFFIN, which stands for "MUlti Frequency image reconstruction For radio INterferometry". MUFFIN estimates the sky images at various frequency bands given the corresponding dirty images and point spread functions. The reconstruction is achieved by minimizing a data fidelity term and joint spatial and spectral sparse analysis regularization terms. It is consequently non-parametric w.r.t. the spectral behaviour of radio sources. MUFFIN algorithm is endowed with a parallel implementation and an automatic tuning of the regularization parameters, making it scalable and well suited for big data applications such as SKA. Comparisons between MUFFIN and the state-of-the-art wideband reconstruction algorithm are provided.

The MUSE Extremely Deep Field: The cosmic web in emission at high redshift

“…This feature is very interesting for detection or denoising approaches operating scale by scale (Starck et al 1998). Such wavelet signal decompositions have been successfully used in astronomy for various applications: for example X-ray source detection (Finoguenov et al 2020), diffuse light study in compact group of galaxies (da Rocha & Oliveira 2005) and the estimation of faint and diffuse radio components (Dabbech et al 2015;Ammanouil et al 2019).…”

Hyperparameter selection for Discrete Mumford-Shah

“…( 7) and let u(z; Θ) a parametric estimator of the underlying ground truth u, depending on some hyperparameters stored in Θ ∈ R L . For ǫ > 0 a Finite Difference step and ∆ = [δ (1) , . .…”

“…Combining SUGAR with a quasi-Newton descent procedure, a fast algorithm was designed to achieve optimal hyperparameters selection for objective functions of the form (1). This strategy, later extended in [17,32] for correlated noise, proved its efficiency for texture segmentation [32], piecewise linear signal denoising [31], and in spatial-spectral deconvolution for large multispectral data [1]. Contributions -This work focuses on the D-MS functional (2) for joint image denoising and contour detection, whose optimization is performed with SL-PAM, a nonconvex alternated minimization scheme, with descent parameters genuinely chosen to ensure fast convergence.…”

Automated Data-Driven Selection of the Hyperparameters for Total-Variation-Based Texture Segmentation

“…A notable exception is [19], in which numerical experiments are run on uncorrelated multi-component data, the components experiencing different noise levels. The noise being assumed independent, this corresponds to a diagonal covariance matrix S = diag(ρ 2 1 , . .…”

“…SUGAR proved its efficiency in the automated hyperparameter selection in a spatial-spectral deconvolution method for large multispectral data corrupted by i.i.d. Gaussian noise [2] Contributions and outline. We propose a Generalized Stein Unbiased GrAdient estimator of the Risk, for the case of Gaussian noise ζ with any covariance matrix S, using the framework of Ordinary Least Squares, that is (5) with W = I P , enabling to manage different noise levels and correlations in the observed data.…”