GAN Assisted Reduction of Misclassification of Noisy Images
Ziyang QIU, Syunta ITO, Takashi Watanabe · The Proceedings of Mechanical Engineering Congress Japan · 2020
One of the problems in the image classification is the misclassification when there are slight noises that are not visible to the human eyes. In this study, GAN-based machine learning is used to correctly classify the images with noises. VGG16 is adopted as the image classifier. By adding small noises to the images, we artificially create noise images that the classification model misclassifies. A noise reduction model based on GAN method is designed and the facility of the model is evaluated.