Abstract: Image denoising is a well-known problem in image processing. Deep networks can achieve state-of-the-art denoising results based on the quality or quantity of training samples. However, Deep networks face performance saturation when the interference of complex noise and few paired training samples. We introduce a grey relational generative adversarial learning method into the image denoising task. To solve the problem of deep network saturation caused by complex noises and the lack of paired training samples, an adversarial learning network is built to learn latent space distribution of noisy images and reconstruct the distribution of clear images. The grey relation analysis is introduced into the network to deal with the uncertainty of noisy images and improve the ability of adversarial learning. This network is optimized by a new loss function that combined the adversary and grey relation. The loss can reasonably measure the difference between the denoised images and clear images. Experiential results show the proposed method obtains the averaged PSNR of 29.67, which is 0.53 higher than the network without grey relation. Extensive experiments demonstrate the superiority of our approach in image denoising.