Image Repair Based on Least Two-Way Generation Against the Network

• Published in: Intelligent Computing Theories and Application Vol. 13394; pp. 739 - 746
• Main Authors: Hu, Juxi, Cheng, Honglin
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Deep learning; Generative adversarial network; Image restoration; Least squares
• ISBN: 9783031138287; 3031138287
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-13829-4_66

With the rapid development of deep learning in the field of artificial intelligence, deep learning has received more and more attention in the application of computer vision technology. Image restoration is an important application in the field of image generation. The missing information in the original image can be filled in and restored through the contextual information and perceptual information of the image. Traditional physical and mathematical image restoration algorithms pay attention to context information but ignore perception information, and use information pixels around the area to be restored to restore the original image. In the face of a single natural landscape and background, the restoration effect is acceptable However, the traditional image restoration method does not perform well on images related to faces and bodies with a large amount of perceptual information. Since then, the researchers have developed image restoration based on a generative confrontation network. A large number of fake images are generated through the generator, the discriminator optimizes the discrimination, and then the part is cropped for restoration. However, the image quality generated by the traditional generative confrontation network is not high, and the phenomenon of gradient disappearance is prone to appear. Image restoration based on the improved generative confrontation network uses the least squares method of generative confrontation network to perform image restoration. The network is better, the repair effect is better, the quality of the generated image is higher, and the gradient is easier to overcome Disappearance. From the perspective of visual effects, it can make people define the results of image restoration as correct, whether it is from the contextual information of the pixel part connected by the color or the perceptual information of the object, it can be judged as the correct image, and the restoration effect is better.