Research On Image Inpainting Algorithm Based On Attention Mechanism

Image completion is an important issue in the field of computer vision,it can be divided into image inpainting and image outpainting.The purpose of image completion is to automatically restore the lost content based on the known content in the image.For regular missing area,the inpainting method has made good progress,while it is still a big challenge for irregular area.Different from the image completed by the previous inpainting method,we conducted research from around irregular image features.Considering the low convolution efficiency when processing spatial position information,attention mechanism is introduced to extract long-distance and irregular image content,Self-attention mechanism and SENet module are able to focus on global information.In addition,attentive normalization is introduced instead of batch normalization to model long-distance relationships on the network.Different from image inpainting,image outpainting has relatively less context in the image center to capture and more content at the image border to predict.Therefore,classical encoder-decoder pipeline of existing methods may not predict the outstretched unknown content perfectly.In this paper,a novel two-stage siamese adversarial model for image extrapolation,named Siamese Expansion Network(Si ENet)is proposed.Specifically,in two stages,a novel border sensitive convolution named adaptive filling convolution is designed for allowing encoder to predict the unknown content,alleviating the burden of decoder.Besides,to introduce prior knowledge to network and reinforce the inferring ability of encoder,siamese adversarial mechanism is designed to enable our network to model the distribution of covered long range feature as that of uncovered image feature.The results on four datasets has demonstrated that our method outperforms existing state-of-the-arts and could produce realistic results.This article conducts derivative experiments in image inpainting,and does a series of image removal to the cloud,similar to an end-to-end image restoration task.Converting remote sensing images with clouds into remote sensing images without clouds is essentially a pixel-to-pixel problem.A spatial attention mechanism is added to the decoder,which transforms the spatial domain information in the picture accordingly,so that the key information can be extracted to realize cloudy removal.