Image Super-Resolution Reconstruction with Dense Residual Attention Module

Deep convolutional neural networks have recently achieved great success in the field of image super-resolution. However, most of the super-resolution methods based on deep neural network do not make full use of the multi-level features which extracted from low-resolution images, and do not pay attention to the high-frequency information which needs to be reconstructed in the image, so the performances are relatively poor. Aiming at these problems, we propose a dense residual attention module to improve the image reconstruction performance. The dense residual attention module proposed in this paper makes full use of low-level image feature, and the channel spatial attention mechanism makes the network pay more attention to the high-frequency information that the image needs to be repaired, and uses the sub-pixel convolution to complete the image. Experiments were carried out on five benchmark datasets Set5, Set14, BSD100, Urban100 and DIV2K100. When the magnification was 4, the PSNR and SSIM are 32.47 dB/0.8986, 29.72 dB/0.8004, 27.73 dB/0.7423, 26.63 dB/0.8030, 29.43 dB/0.9023. Compared with other methods, we obtain the expected results.