Bayesian sparse solutions to linear inverse problems with non-stationary noise with Student-t priors

Abstract

Bayesian approach has become a commonly used method for inverse problems arising in signal and image processing. One of the main advantages of the Bayesian approach is the possibility to propose unsupervised methods where the likelihood and prior model parameters can be estimated jointly with the main unknowns. In this paper, we propose to consider linear inverse problems in which the noise may be non-stationary and where we are looking for a sparse solution. To consider both of these requirements, we propose to use Student-t prior model both for the noise of the forward model and the unknown signal or image. The main interest of the Student-t prior model is its Infinite Gaussian Scale Mixture (IGSM) property. Using the resulted hierarchical prior models we obtain a joint posterior probability distribution of the unknowns of interest (input signal or image) and their associated hidden variables. To be able to propose practical methods, we use either a Joint Maximum A Posteriori (JMAP) estimator or an appropriate Variational Bayesian Approximation (VBA) technique to compute the Posterior Mean (PM) values. The proposed method is applied in many inverse problems such as deconvolution, image restoration and computed tomography. In this paper, we show only some results in signal deconvolution and in periodic components determination of some biological signals related to dynamic circadian clock period determination for cancer studies.

Quantum tomography of electrical currents

“…The sensibility to errors, due to lost information, correspond to zero or close to zero values of the Fourier transform of h n . In order to find a more robust estimation with correct physical properties, we propose to add appropriate prior information on ∆W S,n (ω) thanks to a Bayesian framework [56][57][58] .…”

A High-Resolution and Low-Frequency Acoustic Beamforming Based on Bayesian Inference and Non-Synchronous Measurements

“…To enforce the solution sparsity, there are several kinds [3], [5], [15], [25] of priors such as Double Exponential, Generalized Gaussian, and some other mixture models with heavy tails. In this paper, the Student-t distribution is investigated to promote sparse solutions for two reasons.…”

“…Most of these drawbacks can be overcome by Bayesian inference methods [15], [20], [25]. It can adaptively estimate both unknown random variables and unknown model parameters by applying the Bayes' rule in updating the probability law, in which, a posterior probability can be obtained from the likelihood and prior models.…”

Sparse Bayesian Learning with joint noise robustness and signal sparsity

“…Here we extend the idea of signal-sparsity enforcing mentioned above to the noise-robustness enhancement. That is, instead of a Gaussian noise model, a hierarchical structure-based paradigm is used in noise modelling, so that the induced loss functions have acceptable robustness to heavy-tailed errors [8,27,32]. However, while a variety of fast inference schemes have been developed to improve the practical usefulness of SBL including reducing computation cost and improving convergence rate, these schemes often fail to produce reliable estimates of noise parameters even under the homogeneous Gaussian noise assumption [29]; thus, they have to exclude the estimations of noise parameters because inaccurate estimates of noise parameters can prevent from the success of sparse signal recovery [31].…”

“…Sparse signal recovery is an emerging field in signal processing [1–3]. After introduced in [1, 2], it has been applied to medical imaging [4], channel estimation [5, 6], wireless sensing [7], biological studies [8], to name just a few. Consider the problem of sparse signal recovery with the measurement model given bywhere is an available measurement vector; is a matrix whose columns represent an overcomplete or redundant basis (i.e.…”