Study on salt-and-pepper denoising based on dual convolutional neural networks
Chengqiang Huang, Yinghu He · 2023
Salt-and-pepper denoising based on dual convolutional neural networks (CNNs) is proposed in this paper. Different from the traditional denoising with single CNN, this paper makes a breakthrough from two aspects including noise marking and noise removal, thus CNN for noise marking (MCNN) and CNN for denoising (DCNN) are built. MCNN aims to achieve accurate noise marking, while DCNN aims to achieve high performance denoising. Specifically, the denoising strategy adopted in this paper is not to process the pixel marked as normal by MCNN. For the pixel marked as noise, gray level generated by DCNN is adopted to replace. Experimental results show that the misjudging rate of MCNN marker is reduced by 75.47%, 75.43% and 23.36% compared with the extreme marking, the average marking and the extreme image block marking, respectively. The denoising effect of the dual CNNs proposed in this paper is better than that of the traditional non-CNNs and CNNs. PSNR is 9.53% and 9.42% higher than that of the repeated filter and Chen CNN, and MSE is 21.19% and 28.07% lower than that of the repeated filter and Chen CNN, respectively.