Super-Resolution Reconstruction Method of Face Image Based on Attention Mechanism

Abstract

In recent years, convolutional neural network in Single image super-resolution field show good results. Deep networks can establish complex mapping between low-resolution and high-resolution images, making the reconstructed images quality a great progress over traditional methods. In order to be able to generate face images with rich texture details, the algorithm proposed in this paper captures implicit weight information in channel and space domains through dual attention modules, so as to allocate computing resources more effectively and speed up the network convergence. Fusion of global features through residual connections in this network not only focus on learning the high frequency information of images that has been lost, but also accelerate the network convergence through effective feature supervision. In order to alleviate the defects of MAE loss function, a special Huber loss function is introduced in the algorithm. The experimental results on benchmark show that the proposed algorithm has a significant improvement in image reconstruction accuracy compared with existed SISR methods.