Impulse noise threshold detection and Noise removal using Non-local Grid Network

Abstract

The method of removing the existing Impulse noise was to find the threshold and apply the median filter to remove the noise. In this paper, we try to find reference values in neural networks. In order to find noise, we used a non-local method to consider not only the information of the region but also the entire region. We then utilized Yolo's grid cell structure to make stronger predictions considering various sizes and different areas. Accordingly, qualitative and quantitative evaluation results show better performance than conventional median filters.