Non-local self-similarity recurrent neural network: dataset and study

The images and videos of the high-voltage copper contact are disturbed by various noises in the factory. In this paper, an improved Non-local Self-similarity Recurrent Neural Network(NSRNN) is proposed for image denoising. The sparse representation is used for initializing the images, and then NSRNN is trained and tested based on the image datasets with different noise levels and magnification. Due to the similarity and the time correlation between the sequential images, RNN is used to improve the parameter utilization and model robustness. By measuring the self-similarity of the neighborhood features, NSRNN model outperforms other state-of-the-art methods in term of image denoising performance.