Image Restoration Using Deep Neural Network for Various Degradation

Image signals can be degraded due to various factors such as noise and blurring. Numerous studies have been conducted to restore degraded image signals, and deep learning-based restoration techniques have demonstrated outstanding performance. In this paper, we propose a novel deep neural network architecture designed to restore images degraded by various causes such as noise, Gaussian blur and motion blur. We present improved structures of the widely used U-Net architecture for image restoration. By adding the short cut connections used in ResNet to the conventional U-Net architecture, a new structure that enhances overall convergence performance is proposed. Through extensive computer simulations on various types of degradation and images, we demonstrate that the proposed deep neural network achieves superior performance in restoring images degraded by diverse forms of degradation, compared to conventional deep neural networks.