Abstract: As an important research direction in the field of image processing, image restoration has attracted much attention. The existing image restoration models only focus on one aspect of texture or structure, but ignore that the two complement each other in image restoration tasks, which leads to unsatisfactory images. In this paper, the codecs of double tributaries are used as the backbone of the image generator, corresponding to the texture features and structure features of the generated image respectively, so as to achieve the effect of structure constraint texture and texture guidance structure. Secondly, The bidirectional residual feature fusion module is used to fuse texture features and structure features generated by the decoder to complete the information exchange of the two features. Finally, The detailed features of the restored images are enriched with a multi-scale context feature information aggregation module. Experiments show that the proposed method can improve the SSIM and PSNR values under different mask rates.