Research On Deep Learning Based Image Inpainting Algorithm Guided By Edge Information

Image inpainting is a significant research branch in the field of digital image processing.It refers to the restoration of the defective image through technical means to make it as the same as the original image as much as possible.At present,image inpainting is widely used in medical,security,military,and film industries,so it has very important research value.In recent years,deep learning,especially the emergence of generative adversarial networks,has provided better technologies and methods for image inpainting research.Edge is an important information in digital images,it can reflect the information of the object to a certain extent,and determine the shape and boundary of the object.In image inpainting,there is often a problem that the boundaries of objects and characters in the original image cannot be determined,and the image cannot be restored correctly,so edge information can be used to guide image inpainting.This paper will study the influence of edges guided image inpainting methods based on deep learning,and propose an image inpainting method based on edge prediction and guidance.The method consists of two parts.The first part is a generative adversarial network,which is used to predict and repair the missing edges of the image to be repaired;the second part is an image inpainting network guided by edge information.After comparing with the restoration image without edge guidance,it can be found that the restoration effect is significantly improved.The experimental results show that using edge information to guide the generation of adversarial networks for training can obtain better image inpainting results.The major work of this article can mainly be reflected in the following aspects:1.In this paper,a residual U-Net is used as the generator of the GAN to predict the edges of the missing parts of the image and obtain complete image edge information.Experimental results show that the network can predict most of the key missing edge information.2.The attention module is added to the image repair network,which can effectively improve the training speed after comparison.The predicted edge image is used as the network input for guidance,and the result is compared with the restored image without edge guidance.The results show that a better repair result can be obtained with the addition of edge guidance.3.Use Python language to implement an image restoration system.The system functions include image reading,defacement of any part of the image,restoration of defaced images,prediction of edges,and display of repair results.