Parallel Capsule Network: A Novel Method for Image Denoising

Image denoising is a traditional and hot research topic. During the past years, tremendous advances in image denoising tasks have been achieved using Convolutional Neural Networks (CNN), filtering, sparse coding, classical external priors, or deep learning methods. Although these methods can gain good performance in image denoising field. However, in real life, due to different sources of noise, noise will be generated in several stages such as image acquisition, transmission, and compression. So, we need more about image denoising methods. In this work, we present a novel denoising image method that is based on image classification. To achieve denoising images, the result of image classification can be regarded as prior information for image denoising. We call it the relationship between global and local, such relationship between global and local finds it is a foothold in gestalt psychology, where they believe that the global is not a simple sum or addition of locals, the global is not determined by the locals, and the various locals of the global are determined by the internal structure and nature of the global, so the gestalt organization law means that people always perceive Will organize the empirical materials into a meaningful whole according to a certain form. We evaluated our model on the MNIST dataset and FASHION MNIST dataset. To the best of our knowledge, this is the first such an effort on image denoising in image classification.