Search

The development of deep learning advances the field of image processing. In recent years, lots of methods have made out- standing achievements in the domain of text-to-image synthesis, like Generative Adversarial Networks (GANs). Until now, although some evaluation metrics has been proposed to measure the performance of GANs in text-to-image synthesis, the quality of these evaluation metrics has always been controversial. At present, there is no widely used evaluation metric to judge the quality of generated image. In this paper, a novel No-Reference image quality evaluation metric is proposed, which can be used to get a score for each generated image produced by deep learning without referring to the real image. This evaluation metric can provide a new way to verify the quality of complex networks by judging the quality of generated images retroactively.