Fast video super-resolution using artificial neural networks

In this study, video super-resolution using artificial neural network (ANN) is proposed to enlarge low-resolution (LR) frames. The proposed super-resolution method consists of three main modules, i.e., motion-trace volume collection, ANN training, and ANN prediction. In the proposed method, the LR frames are super-resolved to HR frames through ANN. The traditional motion estimation is used to catch the motion-trace volume which eliminates the unfathomable object motion in the video. Then, the complex spatio-temporal detail between LR and HR data is learned by ANN. Using the ANN training results, the optimal weights can be determined for frame resolution enhancement in video. Simulation results show that the proposed method successfully improves the average peak signal-to-noise ratio (PSNR) and perceptual quality in super-resolved frames.