||Low-complexity 1D-convolutional Neural Network for Super Resolution
||Jangsoo Park;Jongsuk Lee;Donggyu Sim
||Super-resolution; Deep-learning; Low-complexity; CNN
||This paper proposes a method for accelerating deep learning based super-resolution technology. In order to alleviate the complexity of the deep learning based super-resolution technology, the proposed method extracts the horizontal and vertical high-frequency signals separately using one-dimensional filters. Then, the final super-resolution image is obtained by the proposed network from the horizontal and vertical high-frequency signals and the low-resolution input image. The proposed method requires a low computational complexity by using only onedimensional filters due to a smaller number of weights. The proposed method in the high-resolution image restoration experiment shows that the average visual quality in PSNR and SSIM is comparable to ones of VDSR. However, the average speed performance is accelerated by 86.57%.