Residual Dilated U-Net with Spatially Adaptive Normalization for the Restoration of Under Display Camera Images

Recent developments in display and imaging devices have prompted manufacturers to place a camera under the display screen, delivering a larger display-to-body ratio that is genuinely full-screen. However, this Under-Display Camera (UDC) imaging system suffers from severe image degradation such as noise, blur, low-light, and color-shift. This is due to the low light transmission and diffraction property of the display panels. To tackle this issue, we present an end-to-end framework based on U-Net to restore the degraded image. Since the point spread function (PSF) of the UDC degradation is known to be spatially dispersed, we utilize dilated convolutions to increase the receptive field of the model. Furthermore, we use spatially adaptive normalization to regularize feature maps to help restore the image efficiently, thereby improving the performance of our model. We show that our method can restore UDC images with fewer artifacts and produce competitive results to state-of-the-art methods.