可逆去噪网络及其应用研究

Research on Reversible Denoising Network and Its Application

摘要: 近年来，图像处理技术被广泛应用于各个领域。其中，图像去噪部分更是最常被应用于医学、军事等领域。相对于一些传统的去噪方式，虽然在去除合成噪声方面取得较好的成绩，但由于其假设与真实场景下的假设不同，使得在使用传统方法进行现实噪声去噪时，其有效性受到了一定的影响。因此，基于以上因素，本文使用了一种可逆网络去噪方法。同时在可逆网络去噪的基础之上，我们加入了自己的想法进行了一定的改进，在对比试验中，我们选取了SIDD等数据集，进行了对比试验。结果表明：将Haar小波变换改为多贝西小波变换后的图像PSNR有一定的提升，并且加入残差网络预训练后的图像PSNR也有提升。由此可见，我们改进后的可逆去噪网络相较原始去噪网络在图像去噪性能方面有一定的提升。

Abstract: In recent years, image processing technology has been widely used in various fields. Among them, the image denoising part is most often used in medical, military and other fields. Compared with some traditional denoising methods, although it has achieved good results in removing synthetic noise, its effectiveness is affected by the use of traditional methods for real noise denoising due to the different assumptions from the real scene. Therefore, based on the above factors, a reversible network denoising method is used in this paper. At the same time, on the basis of reversible net-work denoising, we added our own ideas to make some improvements. In the comparison test, we selected SIDD and other data sets for comparison test. The results show that the PSNR of the image after changing the Haar wavelet transform to the Dobbercy wavelet transform has a certain im-provement, and the PSNR of the image after adding the residual network pre-training is also im-proved. It can be seen that our improved reversible denoising network has a certain improvement in image denoising performance compared with the original denoising network.