Acceleration of RED via Vector Extrapolation

Add time:08/02/2019    Source:sciencedirect.com

Models play an important role in inverse problems, serving as the prior for representing the original signal to be recovered. REgularization by Denoising (RED) is a recently introduced general framework for constructing such priors using state-of-the-art denoising algorithms. Using RED, solving inverse problems is shown to amount to an iterated denoising process. However, as the complexity of denoising algorithms is generally high, this might lead to an overall slow algorithm. In this paper, we suggest an accelerated technique based on vector extrapolation (VE) to speed-up existing RED solvers. Numerical experiments validate the obtained gain by VE, leading to a substantial savings in computations compared with the original fixed-point method.

We also recommend Trading Suppliers and Manufacturers of Basic Red 22 (cas 12221-52-2). Pls Click Website Link as below: cas 12221-52-2 suppliers

Prev:Study on the changes of emissivity of basic copper carbonate in the decomposition process
Next:Neutron noise analysis using the basic element method)