I’ll describe a general framework for rapidly learning image processing operators, applying them, and inverting them. We developed the first application of this approach in 2016: the Rapid and Accurate Image Super Resolution (RAISR) method for fast and high quality image upscaling, using a trained set of filters. I will detail this work first, which has since been used in a number of products at Google including the Pixel 2 phones. Then, I will describe a broad generalization of RAISR, a more powerful trainable image-adaptive filtering framework that is still easy to train, computationally efficient, and useful for a wide range of problems in computational photography and imaging. I will illustrate applications to a variety of operations which may appear in a camera pipeline and elsewhere, including denoising, demosaicing, and stylization. Finally, I will describe how such a framework can also be used to solve various other inverse problems very efficiently.
This presentation is part of Minisymposium “MS50 - Analysis, Optimization, and Applications of Machine Learning in Imaging (3 parts)”
organized by: Michael Moeller (University of Siegen) , Gitta Kutyniok (Technische Universität Berlin) .