Comparison and analysis of classical image denoising methods based on convolution neural network
Hongyu Liu, Hua Qu · 2022
Nowadays, internet and communication are well developed and all parts of the world are interconnected. People receive a lot of data all the time. According to statistics, visual information accounts for more than 60% of the information received by human eyes. As a part of machine learning, deep neural network has greatly promoted the development of artificial intelligence. Due to the imperfection of the equipment and system, the image loses some features due to the influence of noise during the transmission process, which makes the image unclear, which not only affects the normal recognition of people, but also affects the subsequent segmentation, extraction, detection and recognition of the image. At this time, it is necessary to take measures to make the image clearer. Image denoising refers to the process of eliminating the noise in the digital image as much as possible. In short, it is not only to remove the noise pollution, but also to retain the useful feature information of the image itself as much as possible. In addition to improving the image quality, it can also better identify the image. This paper is a research on image denoising methods based on convolution neural network (CNN). Using neural network’s good learning ability of image statistical characteristics, four CNN based methods of DnCNN, FFDnet, CBDnet and SRMD are used to realize image denoising. It is found that DnCNN has the best denoising effect on Gaussian noise, followed by FFDnet and SRMD.