Abstract
Although video super resolution(VSR) is one of the essential technologies in the area of video processing using deep neural network(DNN), it is difficult for the VSR network to utilize the temporal correlation between consecutive input video frames. Recently, convolutional neural networks(CNN) based VSR methods show a significant improvement to generate high resolution(HR) videos from low resolution(LR) videos. In this paper, we propose a VSR method using deformable convolution based alignment network with residual dense block. It enables to the proposed network to use the correlated information from the intermediate feature maps for the purpose of improving the quality of VSR. Compared to the previous methods, experimental results show that the proposed method achieves both PSNR and SSIM by as much as 0.23dB and 0.006 on average, respectively.