Abstract: Aiming at the problem of low image resolution caused by internal and external factors such as face posture， uneven lighting， shooting environment， and shooting equipment， this paper presents a high-resolution face recognition image model combined with attention mechanism. Firstly， low-resolution face image pairs are used as input to the two generators， and facial feature information is extracted through the residual block and attention module stacking network， and then high-resolution people are generated. Face image. A discriminator is used in training to supervise the training process of the two generators. The Adam algorithm is used to iteratively optimize the discriminator， generator and counter loss function to improve the performance of the network model. The model is trained on the CASIA-WebFace dataset and tested on the CASIA-WebFace， CelebA， and LFW datasets. Experiments show that the attention module can effectively compensate for the lack of global information learning due to the partial relationship modeling when the shallow convolutional neural network extracts features， and it can learn the feature information that is conducive to the reconstruction of high-resolution images. This model has a good visual effect in reconstructing the face， and has the characteristics of retaining the identity information of the face.