Application Of Generative Adversarial Networks In Image Completion

Image Completion is an important research direction in the field of digital image processing.It is a technique for filling alternative content in the missing area of the image.Using this technique,the missing areas in the image can be complemented according to certain rules,so that the completed image achieves a false effect.It is the perfect complementing effect that makes the image completion technology have broad application prospects in the restoration of ancient cultural relics and calligraphy,the diagnostic aid of modern medical treatment and the image acquisition of remote sensing satellites.The existing mature image completion methods include: diffusion-based image completion method,sample block-based image completion method,and image completion method based on structural features.However,these image completion methods have their own shortcomings: the diffusion-based method can only fill the missing area with small area or narrow area,and the effect of complementing the area is not ideal;the method based on sample block relies too much In the intact area of the image,if there is no segment similar to the image missing area in the intact area of the image,the missing area of the image will not be well complemented;the structure-based method improves the image by retaining important structural features.The effect of completion,however,is limited to specific structures and is not universal.In view of the many shortcomings of existing image completion methods,people try to introduce the relevant theory of deep learning into the field of image completion to improve the image completion effect.This paper is also inspired by this and proposes an image completion method based on Generative adversarial networks(GAN).In this method,the generator-oriented network model consists of two parts: the generator model and the discriminator model,all of which use the Convolutional Neural Network(CNN)to achieve its basic functions.The method completes the missing area of the image by the generator model and uses the discriminator model to discriminate the image complementing effect.In order to make the complemented image more reasonable in detail processing and to make the pixels in the completion region maintain continuity with the surrounding pixels on the image feature with the maximum probability,the method adopts Markov random field(MRF).A loss function training generator model combined with a Mean square error(MSE).The peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)image evaluation criteria are used to evaluate the complementing effect of the image completion method proposed in this paper.The evaluation results show that the image completion method based on the Generative adversarial networks has better completion effect than other existing methods.