Robust Bayesian method for simultaneous block sparse signal recovery with applications to face recognition

Abstract

In this paper, we present a novel Bayesian approach to recover simultaneously block sparse signals in the presence of outliers. The key advantage of our proposed method is the ability to handle nonstationary outliers, i.e. outliers which have time varying support. We validate our approach with empirical results showing the superiority of the proposed method over competing approaches in synthetic data experiments as well as the multiple measurement face recognition problem.

An artificial bee colony optimization based matching pursuit approach for ultrasonic echo estimation

“…The existing SSR algorithms are commonly developed by simplifying the NP-hard problem as a constrained optimization problem by greedy approximations or by applying L 1 -norm or L p -norm constraints on the decomposition coefficients to find sub-optimal solutions. Many SSR algorithms have been developed in the past two decade, such as matching pursuit (MP) [4], greedy basis pursuit [5], Sparse Bayesian learning (SBL) [6], nonconvex regularization [7], and applications of SSR extend into many fields [8][9][10][11][12].…”

Relevance Subject Machine: A Novel Person Re-identification Framework

“…RVM has been shown to achieve much higher SR rates than many deterministic approaches [31] and has the potential to outperform algorithms based on the formulation in (1) in the re-id task. We build upon the signal model we introduced in [32] and refer to our approach as the Relevance Subject Machine (RSM).…”

“…The combination of block sparsity, robustness to sparse noise, and joint sparsity over multiple observations was first studied by us in [32]. The signal model considered was…”

Generalized Residual Ratio Thresholding

“…This letter considers the recovery 1 of high dimensional structured sparse signals from low dimensional linear measurements, a problem relevant in many signal processing and machine learning applications [1]- [4]. This letter considers two structured sparse recovery scenarios given by a) multiple measurement vector (MMV) model and b) block sparse (BS) model.…”

“…In BS, we consider a regression model with L = 1. However, the p entries of B are divided into p b = p/l b nonoverlapping blocks of equal size l b such that the entries in 1 The following notations are used. X[i, j] denotes the (i, j) th entry of a matrix X. X[:, I] and X[I, :] denote the columns and rows of matrix X indexed by I. X T , X −1 and X † represent the transpose, inverse and Moore-Penrose pseudo inverse of X respectively.…”