IS20: SP1: Early Career Prize Lecture - Regularization by Denoising (RED) скачать в хорошем качестве

IS20: SP1: Early Career Prize Lecture - Regularization by Denoising (RED)

Friday, July 10 11:00 AM - 11:45 AM Image denoising is the most fundamental problem in image restoration, and it is considerably solved: it has reached impressive heights in performance and quality — almost as good as it can ever get. But interestingly, it turns out that we can solve many other problems using the image denoising “engine.” This talk presents the Regularization by Denoising (RED) framework: using a denoising engine in formulating the regularization of any inverse problem. The idea is to define an explicit image-adaptive regularization functional by using a high-performance denoiser. Under some conditions, we will see that the resulting regularizer is guaranteed to be convex, and the overall objective function is explicit, clear, and well-defined. With complete flexibility to choose the iterative optimization procedure for minimizing this functional, RED is capable of incorporating any image denoising algorithm as a regularizer and treat general inverse problems very effectively. Lastly, we will discuss recent advances as well as limitations and future challenges that arise from them. Yaniv Romano Stanford University, U.S.