Abstract:

Because of the limitation of the imaging equipment, the face images captured by it usually have the problem of low resolution and low quality. This paper proposes a method based on the generative adversarial network and attention mechanism for the multi-scale super-resolution of face images. In this paper, the deep residual network and the deep convolutional neural network (VGG-net) are used as the generator and the discriminator, respectively. The attention modules are combined with the residual blocks in the deep residual network to reconstruct face images which are highly similar to the high-resolution images and difficult for the discriminator to distinguish. Experimental results demonstrate the effectiveness of the proposed method in multi-scale face image super-resolution and the important role of the attention mechanism in image detail reconstruction.

Key words: super-resolution, generative adversarial network, attention mechanism, deep residual network, deep convolutional neural network