Understanding of "image denoising using an improved generic advantageous network with Wasserstein distance"

translate ：" be based on Wasserstein The improvement of distance generates anti network image denoising "

-- Proceedings of the 40th Chinese Control Conference -- 2021

---

One 、 Generalization

        This article proposes a method based on WGAN-GP Image denoising algorithm based on . This method can directly generate clean images from noisy images , And can well retain the texture details of the image . utilize WGAN-GP Loss To fight the loss , Avoid the gradient disappearing during training , Make the training more stable , And add Loss of perception To improve the quality of the generated image , Reduce the cost of Computing .

Two 、 Method

2.1、 Noise reduction model

        First , The purpose of this model is to find a G, Will degrade the image z Map to a clean image x：

2.2、 Loss function

        The losses are as follows ：

  among ：

2.3、 Network structure

        The structure is as follows ：

 

3、 ... and 、 experimental result