Ilker Bayram

Sparse Signal Estimation by Maximally Sparse Convex Optimization

This paper addresses the problem of sparsity penalized least squares for applications in sparse signal processing, e.g., sparse deconvolution. This paper aims to induce sparsity more strongly than L1 norm regularization, while avoiding non-convex optimization. For this purpose, this paper describes the design and use of non-convex penalty functions (regularizers) constrained so as to ensure the convexity of the total cost function F to be minimized. The method is based on parametric penalty functions, the parameters of which are constrained to ensure convexity of F. It is shown that optimal parameters can be obtained by semidefinite programming (SDP). This maximally sparse convex (MSC) approach yields maximally non-convex sparsity-inducing penalty functions constrained such that the total cost function F is convex. It is demonstrated that iterative MSC (IMSC) can yield solutions substantially more sparse than the standard convex sparsity-inducing approach, i.e., L1 norm minimization.

On the Dual-Tree Complex Wavelet Packet and

The two-band discrete wavelet transform (DWT) provides an octave-band analysis in the frequency domain, but this might not be ldquooptimalrdquo for a given signal. The discrete wavelet packet transform (DWPT) provides a dictionary of bases over which one can search for an optimal representation (without constraining the analysis to an octave-band one) for the signal at hand. However, it is well known that both the DWT and the DWPT are shift-varying. Also, when these transforms are extended to 2-D and higher dimensions using tensor products, they do not provide a geometrically oriented analysis. The dual-tree complex wavelet transform , introduced by Kingsbury, is approximately shift-invariant and provides directional analysis in 2-D and higher dimensions. In this paper, we propose a method to implement a dual-tree complex wavelet packet transform , extending the as the DWPT extends the DWT. To find the best complex wavelet packet frame for a given signal, we adapt the basis selection algorithm by Coifman and Wickerhauser, providing a solution to the basis selection problem for the . Lastly, we show how to extend the two-band to an -band (provided that ) using the same method.

M

Total variation (TV) denoising is an effective noise suppression method when the derivative of the underlying signal is known to be sparse. TV denoising is defined in terms of a convex optimization problem involving a quadratic data fidelity term and a convex regularization term. A non-convex regularizer can promote sparsity more strongly, but generally leads to a non-convex optimization problem with non-optimal local minima. This letter proposes the use of a non-convex regularizer constrained so that the total objective function to be minimized maintains its convexity. Conditions for a non-convex regularizer are given that ensure the total TV denoising objective function is convex. An efficient algorithm is given for the resulting problem.

-Band Transforms

Total variation (TV) is a powerful method that brings great benefit for edge-preserving regularization. Despite being widely employed in image processing, it has restricted applicability for 1-D signal processing since piecewise-constant signals form a rather limited model for many applications. Here we generalize conventional TV in 1-D by extending the derivative operator, which is within the regularization term, to any linear differential operator. This provides flexibility for tailoring the approach to the presence of nontrivial linear systems and for different types of driving signals such as spike-like, piecewise-constant, and so on. Conventional TV remains a special case of this general framework. We illustrate the feasibility of the method by considering a nontrivial linear system and different types of driving signals.

Convex 1-D Total Variation Denoising with Non-convex Regularization

This paper develops an overcomplete discrete wavelet transform (DWT) based on rational dilation factors for discrete-time signals. The proposed overcomplete rational DWT is implemented using self-inverting FIR filter banks, is approximately shift-invariant, and can provide a dense sampling of the time-frequency plane. A straightforward algorithm is described for the construction of minimal-length perfect reconstruction filters with a specified number of vanishing moments; whereas, in the nonredundant rational case, no such algorithm is available. The algorithm is based on matrix spectral factorization. The analysis/synthesis functions (discrete-time wavelets) can be very smooth and can be designed to closely approximate the derivatives of the Gaussian function.

A Signal Processing Approach to Generalized 1-D Total Variation

This paper introduces a “directional total variation” (TV) where the gradients are weighted depending on their direction. The introduced directional TV has increased (and tunable) sensitivity to variations at a selected direction. In order to demonstrate the utility of the directional TV, we consider an image denoising formulation. This formulation requires the realization of the “proximal map” of the directional TV. Therefore, it is relevant for more general inverse problem settings as well. We derive an algorithm that solves the problem and use the algorithm to study the effects of the parameters of the directional TV.

Overcomplete Discrete Wavelet Transforms With Rational Dilation Factors

In a number of signal processing applications, problem formulations based on the ℓ1 norm as a sparsity inducing signal prior lead to simple algorithms with good performance. However, ℓ1 norm is not flexible enough to handle certain signal structures that are represented using a few groups of coefficients. Formulations that make use of mixed norms provide an alternative that can handle such signals by forcing sparsity on a group level and allowing non-sparse distributions within the groups. However, conventional mixed norms allow only non-overlapping groups - a restriction that leads to characteristics unlikely for natural signals. In this paper, we investigate mixed norms with overlapping groups. We consider a simple denoising formulation that gives a convex optimization problem and provide an algorithm that solves the problem. We use the algorithm to evaluate the performance of mixed norms with overlapping groups as signal priors.

A directional total variation

We investigate a subband adaptive version of the popular iterative shrinkage/thresholding algorithm that takes different update steps and thresholds for each subband. In particular, we provide a condition that ensures convergence and discuss why making the algorithm subband adaptive accelerates the convergence. We also give an algorithm to select appropriate update steps and thresholds for when the distortion operator is linear and time invariant. The results in this paper may be regarded as extensions of the recent work by Vonesch and Unser.

Mixed norms with overlapping groups as signal priors

We develop a rational-dilation wavelet transform for which the dilation factor, the Q-factor and the redundancy can be easily specified. The introduced transform contains Hilbert transform pairs of atoms, therefore it is also suitable for oscillatory signal processing. The transform may be modified to obtain a tight chirplet frame for discrete-time signals. A fast implementation, that makes use of an equivalent filter bank, makes the transform suitable for long signals. Examples on natural signals are provided to demonstrate the utility of the transform.

A Subband Adaptive Iterative Shrinkage/Thresholding Algorithm

In this letter, we expand upon the method of Tay for the design of orthonormal ldquoQ-shiftrdquo filters for the dual-tree complex wavelet transform. The method of Tay searches for good Hilbert-pairs in a one-parameter family of conjugate-quadrature filters that have one vanishing moment less than the Daubechies conjugate-quadrature filters (CQFs). In this letter, we compute feasible sets for one- and two-parameter families of CQFs by employing the trace parameterization of nonnegative trigonometric polynomials and semidefinite programming. This permits the design of CQF pairs that define complex wavelets that are more nearly analytic, yet still have a high number of vanishing moments.