Single-Image Super-Resolution Reconstruction Based on the Differences of Neighboring Pixels

The deep learning technique was used to increase the performance of single image super-resolution (SISR). However, most existing CNN-based SISR approaches primarily focus on establishing deeper or larger networks extract more significant high-level features. Usually, pixel-level loss between target high-resolution and estimated is used, but neighbor relations pixels in are seldom used. On other hand, according observations, a pixel's relationship contains rich information about spatial structure, local context, structural knowledge. Based this fact, paper, we utilize relationships different perspective, propose differences neighboring regularize CNN by constructing graph from ground-truth image. proposed method outperforms state-of-the-art methods terms quantitative qualitative evaluation benchmark datasets. Keywords: Super-resolution, Convolutional Neural Networks, Deep Learning