In image processing applications, image intensity is often measured via the counting of incident photons emitted by the object of interest. In such cases, image data-noise is accurately modeled by a Poisson distribution. This motivates the use of Poisson maximum likelihood estimation for image reconstruction. However, when the underlying model equation is ill-posed, regularization is needed. Regularized Poisson likelihood estimation has been studied extensively by the authors, though a problem of high importance remains: the choice of the regularization parameter. We will present three statistically motivated methods for choosing the regularization parameter, and numerical examples will be presented to illustrate their effectiveness.

In numerous applications of image processing, e.g. astronomical and medical imaging, data-noise is well-modeled by a Poisson distribution. This motivates the use of the negative-log Poisson likelihood function for data fitting. (The fact that application scientists in both astronomical and medical imaging regularly choose this function for data fitting provides further motivation.) However difficulties arise when the negative-log Poisson likelihood is used. Chief among them are the facts that it is non-quadratic and is defined only for vectors with nonnegative values. The nonnegatively constrained, convex optimization problems that arise when the negative-log Poisson likelihood is used are therefore more challenging than when least squares is the fit-to-data function. Edge preserving deblurring and denoising has long been a problem of keen interest in the image processing community. While total variation regularization is the gold standard for such problems, its use yields computationally intensive optimization problems. This motivates the desire to develop regularization functions that are edge preserving, but are less difficult to use. We present one such regularization function here. This function is quadratic, and can be viewed as the discretization of a diffusion operator with a diffusion function that is approximately 1 in smooth regions of the true image and is less than 1 (but still positive) at or near an edge. Combining the negative-log Poisson likelihood function with this quadratic, edge preserving regularization function yields a strictly convex, nonnegatively constrained optimization problem. A large portion of this paper is dedicated to the presentation of and convergence proof for an algorithm designed for this problem. Finally, we apply the algorithm to synthetically generated data in order to test the methodology.

In positron emission tomography, image data corresponds to measurements of emitted photons from a radioactive tracer in the subject. Such count data is typically modeled using a Poisson random variable, leading to the use of the negative-log Poisson likelihood fit-to-data function. Regularization is needed, however, in order to guarantee reconstructions with minimal artifacts. Given that tracer densities are primarily smoothly varying, but also contain sharp jumps (or edges), total variation regularization is a natural choice. However, the resulting computational problem is quite challenging. In this paper, we present an efficient computational method for this problem. Convergence of the method has been shown for quadratic regularization functions and here convergence is shown for total variation regularization. We also present three regularization parameter choice methods for use on total variation-regularized negative-log Poisson likelihood problems. We test the computational and regularization parameter selection methods on two synthetic data sets.

Penalized maximum likelihood methods are commonly used in positron emission tomography (PET) and single photon emission computed tomography (SPECT). Due to the fact that a Poisson data-noise model is typically assumed, standard regularization parameter choice methods, such as the discrepancy principle or generalized cross validation, cannot be directly applied. In the recent work of the authors, regularization parameter choice methods for penalized negative-log Poisson likelihood problems are introduced. In this article, we apply these methods to the applications of PET and SPECT, introducing a modification that improves the performance of the methods. We then demonstrate how these techniques can be used to choose the hyper-parameters in a Bayesian hierarchical regularization approach.