一种改进的深度网络残差学习的图像降噪方法

近年来，基于深度卷积神经网络的学习方法在图像降噪方面取得了前所未有的成果，通过调整网络结构和参数来获取更好的图像降噪效果已成为研究热点.降噪卷积神经网络在深度神经网络中采用残差学习方法，在提高降噪效果的同时，在一定程度上解决了盲降噪问题.其不足之处在于算法收敛时间长.该文针对降噪卷积神经网络结构做了进一步的改进，提出了一种基于反卷积降噪神经网络的图像降噪算法.该文工作的主要特色如下：1) 在原有的网络结构中，引入反卷积神经网络，优化了残差学习方式；2) 提出一种新的损失函数计算方法.使用BSD68和SET12测试数据集对本文提出的方法进行验证，实验结果表明，该文算法的降噪性能与降噪卷积神经网络算法相比，在相同降噪效果情形下，该文算法的收敛时间缩短了120%～138%.同时，与传统的深度学习图像降噪算法比较，该文方法的降噪效果和运行效率也都有提高.

An improved image denoising method of depth network residual learning

In recent years， learning methods based on deep convolutional neural network have made unprecedented achievements in image noise reduction. Adjusting network structure and parameters to obtain better image noise reduction effect has become a research hotspot. Denoising convolutional neural networkadopts residual learning method in deep neural network， which solves the problem of blind denoising to some extent while improving the effect of denoising. Its shortcomings is that the algorithm convergence time is long. In this paper， an image denoising algorithm based on deconvolution denoising neural networkis proposed. The main features of this paper are as follows. 1) In the original network structure， deconvolution neural network is introduced to optimize the residual learning mode. 2) A new loss function calculation method is proposed. BSD68 and SET12 test data sets are used to verify the method proposed in this paper. Experimental results show that the denoising performance of the algorithm in this paper is 120%-138% shorter than that of denoising convolutional neural network algorithm under the same denoising effect. At the same time， compared with the traditional deep-learning image denoising algorithm， the denoising effect and operation efficiency of this method are also greatly improved.