STM Topography Image Noise Modeling And Blind Denoising Based On Deep Learning

Scanning tunneling microscope (STM) is an important instrument for atomic-scale surface research, but its images are susceptible to interference. Periodic noise is one of the most common STM image defect. Due to the randomness of noise in STM images, previous repair methods require image prior and classification, and then use special algorithms to denoising. In this paper, a novel STM blind denoising method is proposed. Generative Adversarial Network (GAN) is used to simulate the noise and construct a pair of data sets, and then Convolutional Neural Network (CNN) is used for denoising. The network can realize the blind denoising of STM noise image without image prior and classification. The proposed method has demonstrated superior denoising effectiveness on STM images compared to other conventional algorithms, as evidenced by ample experimental results.