Research On Multiple Image Inpainting And Semantic Integrity Translation Based On Generative Adversary Network

As one of the most popular methods in the field of data generation,automatic generation of deep images is not valuable in the field of real world data generation.This paper intends to start from two basic directions of image processing: image inpainting and translation based on GAN.The details are as follows:Aiming at the problem that general image inpainting can only produce a single result,this paper proposes a noise enhancement based generation countermeasure image inpainting method to achieve image diversity restoration.Specifically,a specific noise is added to the input image,and a two-stage depth regression network is used to enhance the influence of noise,combined with the generation of confrontation network to achieve image diversity restoration.The experimental results show that the proposed method can guarantee the diversity of the results and the rationality of the image structureIn order to solve the problem of semantic loss in some image translation methods,this paper proposes a method based on self-attention mechanism and semantic loss to generate a more plausible image translation result.Specifically,the autoencoder structure is used to extract the target features in the image,the self-attention mechanism is used to enhance the quality of the generated image,and the semantic loss is introduced to improve the correlation between the original image and the target image features.The experimental results show that the proposed method can not only obtain good translation results,but also maintain the semantic integrity of salient objects in the image to a certain extent.