ProxImaL: Efficient Image Optimization using Proximal Algorithms

摘要

Computational photography systems are becoming increasinglydiverse, while computational resources—for example on mobileplatforms—are rapidly increasing. As diverse as these camera systemsmay be, slightly different variants of the underlying imageprocessing tasks, such as demosaicking, deconvolution, denoising,inpainting, image fusion, and alignment, are shared between all ofthese systems. Formal optimization methods have recently beendemonstrated to achieve state-of-the-art quality for many of theseapplications. Unfortunately, different combinations of natural imagepriors and optimization algorithms may be optimal for different problems,and implementing and testing each combination is currentlya time-consuming and error-prone process. ProxImaL is a domainspecificlanguage and compiler for image optimization problemsthat makes it easy to experiment with different problem formulationsand algorithm choices. The language uses proximal operators asthe fundamental building blocks of a variety of linear and nonlinearimage formation models and cost functions, advanced image priors,and noise models. The compiler intelligently chooses the bestway to translate a problem formulation and choice of optimizationalgorithm into an efficient solver implementation. In applicationsto the image processing pipeline, deconvolution in the presence ofPoisson-distributed shot noise, and burst denoising, we show thata few lines of ProxImaL code can generate highly efficient solversthat achieve state-of-the-art results. We also show applications tothe nonlinear and nonconvex problem of phase retrieval.