Image completion is an important research area in digital image processing, and image completion integrating multi-scale information is a hot research topic in recent years. However, most existing works are based on convolutional neural networks to extract image features without considering the multi-scale spatial structure information of the original image, and it's difficult effectively utilize local features and global information. Therefore, this paper proposes a multi-scale deep image completion model (CapsNet-GL) by fusing capsule networks. By combining the Globally and Locally Consistent Image Completion (GL) algorithm, firstly, a Mask region is randomly generated on the real image and input to the complementary network for feature extraction and reconstruction. Then, Capsule Network (CapsNet) is used to improve the global discriminator of GL to obtain richer multi-scale information. Subsequently, the complementation results are input to the global discriminator and the local discriminator respectively to fuse the local information with the global information. Finally, the experimental results on the CelebA-HQ dataset show that the image quality generated by the model proposed in this paper is better in terms of content and structure, and can effectively obtain multi-scale image information and ensure the reliability of the image completion results and their consistency with the original image.