Efficient Structure of Deep Neural Network for Smart Phone Image Super-Resolution

Abstract

Super-Resolution is the process of generating a high-resolution image from a low-resolution input image. In recent years, the super resolution using the deep neural network technique has shown superior results. We experimented with various reconstruction networks to be applied to smartphone images. The convolution layer, the deconvolution layer, the bilinear upsampling, the pixel shuffling, and the bilinear additive upsampling method were experimented for reconstruction part of superresolution network. In the quantitative evaluation, the combination of bi-linear upsampling and convolution has shown the best result.