Image denoising based on improved RDN algorithm
Jiacheng Zhang, Jiaxiu Zhang, Haihui Hong · 2022 IEEE 5th International Conference on Automation, Electronics and Electrical Engineering (AUTEEE) · 2022
In recent years, good results have been achieved in the research of convolution neural networks. However, the image denoising algorithm based on CNN still faces serious challenges, such as poor detail reconstruction ability, many parameters, difficult training, etc. On the basis of the traditional residual dense network (RDN), combined with attention mechanism (CBAM), it can not only effectively alleviate the gradient disappearance caused by the increase of network depth, but also provide a more dense connection mode to realize the reuse of features. Moreover, CBAM can effectively improve the connection of various features in channel and space, and use dense jump connection and residual structure to solve difficult training problems and low feature utilization in information extraction networks. In the experiment, we used aluminum pictures as the data set and added different noise levels to them, compared with the traditional filtering algorithm and deep learning filtering method DNCNN, and to verify its effectiveness. The results show that this method is superior to many other image denoising algorithms in terms of noise reduction and computational complexity.