Image Quality Improve by Super Resolution Generative Adversarial Networks

Abstract

In this research, we analyze the methods to improve the quality of the pictures. Machine learning technique is used to achieve the effect of converting low-pixeled pictures into high pixeled ones. It is convenient and can be used in many circumstances, such as engineering projects and medical surveys. This study uses the SRGAN to train the model, which has the generator network and the discriminator network. The generator network is used to generate high-resolution images, and the discriminator network is used to judge the authenticity of the image generated. Our network has trained the network with 8× upscaling factors and eventually obtains predominant achievement for dealing with detailed textures like trees, cars, and animal fur. Our network can recover low-quality pictures perpetual satisfying, and the loss of this model is low. Our research provides model-based strategic support for image quality improvement and gets a good picture resolution increase in the tests.