The Equivalence of Half-Quadratic Minimization and the Gradient Linearization Iteration

Abstract

A popular way to restore images comprising edges is to minimize a cost-function combining a quadratic data-fidelity term and an edge-preserving (possibly nonconvex) regularization term. Mainly because of the latter term, the calculation of the solution is slow and cumbersome. Half-quadratic (HQ) minimization (multiplicative form) was pioneered by Geman and Reynolds (1992) in order to alleviate the computational task in the context of image reconstruction with non-convex regularization. By promoting the idea of locally homogeneous image models with a continuous-valued line process, they reformulated the optimization problem in terms of an augmented cost function which is quadratic with respect to the image and separable with respect to the line process. Hence the name "half quadratic". Since then, a large amount of papers were dedicated to HQ minimization and important results including edge-preservation along with convex regularization and convergence have been obtained.In this paper we show that HQ minimization (multiplicative form) is equivalent to the most simple and basic method where the gradient of the cost-function is linearized at each iteration step. In fact, both methods give exactly the same iterations. Furthermore, connections of HQ minimization with other methods, such as the quasi-Newton method and the generalized Weiszfeld's method, are straightforward.

Anisotropic Interpolation of Sparse Generalized Image Samples

“…In this section, we review a class of such functionals described in [26], [29] and defined in our framework as…”

“…Consequently, the basic steepest-descent method applied to the given cost functional is equivalent to the corresponding gradient flow. Similarly, the iteratively reweighted leastsquares (IRLS) technique [25] that is used for nonquadratic regularization is associated with linearized versions of the gradient of the original functional [21], [26]. This provides an interpretation that relates IRLS to lagged-diffusivity fixedpoint iterations.…”

Mathematical Modeling of Textures: Application to Color Image Decomposition with a Projected Gradient Algorithm

“…This method has on the one hand the advantage of being very easy to implement, and on the other hand the disadvantage of being quite slow. To improve the convergence speed, quasi-Newton methods have been proposed [1,23,29,36,55,56,67]. Iterative methods have proved successful [16,18,35].…”

Semi-calibrated Near-Light Photometric Stereo

“…Yet, the practical efficiency of the overall algorithm hinges on the choices of the scaling matrices H (k,k) θ and H (k+1,k) z and the subroutines for solving (13) and (15). In this work, our choices for H (k,k) θ and H (k+1,k) z (see formulas (18) and (21) below) will be structured Hessian approximations motivated from the iteratively reweighted least-squares (IRLS) method [13,14].…”

“…In this work, our choices for H (k,k) θ and H (k+1,k) z (see formulas (18) and (21) below) will be structured Hessian approximations motivated from the iteratively reweighted least-squares (IRLS) method [13,14]. The solution of the resulting subproblem can be interpreted as a regularized Newton step; see [15,16,17].…”