3D Poisson microscopy deconvolution with Hessian Schatten-norm regularization

Inverse problems with shot noise arise in many modern biomedical imaging applications. The main challenge is to obtain an estimate of the underlying specimen from measurements corrupted by Poisson noise. In this work, we propose an efficient framework for photon-limited image reconstruction, under a regularization approach that relies on matrix-valued operators. Our regularizers involve the Hessian operator and its eigenvalues. They are second-order regularizers that are well suited to biomedical images. For the solution of the arising minimization problem, we propose an optimization algorithm based on an augmented-Lagrangian formulation and specifically tailored to the Poisson nature of the noise. To assess the quality of the reconstruction, we provide experimental results on 3D image stacks of biological images for microscopy deconvolution.