Abstract: The popularity of full screens puts forward high-quality under-screen shooting requirements for smartphone front cameras. The current Organic Light-Emitting Diode(OLED) transparent screens used in under-screen shooting schemes have suffered the problems such as light diffraction and refraction, resulting in blurred RGB（Red Green Blue）images and loss of details. Aiming at the problems mentioned above, a method to restore the high-quality RGB image under smart phone-based OLED screen has been presented in this paper. Firstly, we design a novel image acquisition device under smart phone-based OLED transparent screen displays. The collected image data set consists of more than 10 000 typical scene pairs, which is substantially useful in the deep learning based under-screen image restoration. Secondly, an image restoration scheme for under smart phone-based OLED screen is presented by using Generative Adversarial Nets (GAN). In particular, the residual networks in the generator are used to learn the details of the under-screen image, and the perceptual loss function includes color loss, confrontation loss and content loss. The experiments have shown that our method has achieved better qualitatively visual scenes and quantitatively evaluation results based on Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS). And also, our real-data set is proved to be practicable for addressing real complicated image restoration problem.