A Survey on Improved GAN based Image Inpainting

• Published in: 2022 2nd International Conference on Consumer Electronics and Computer Engineering (ICCECE) pp. 319 - 322
• Main Authors: Weng, Yu, Ding, Shiyu, Zhou, Tong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.01.2022
• Subjects: Analytical models; Computer vision; Deep learning; Generative Adversarial Network; Generative adversarial networks; Image Inpainting; Image restoration; Solid modeling; Virtual reality
• DOI: 10.1109/ICCECE54139.2022.9712740

Image inpainting is a computer vision technique that originated from the restoration of broken artworks. It refers to the work of restoring or completing an image by using the information of the undamaged part of the image when there is a broken or empty local area in the image. Image inpainting technology has a wide range of applications in digital cultural heritage preservation, image editing, virtual reality and other fields. This paper introduces the traditional image inpainting methods briefly at first, and then focuses on the inpainting methods based on generative adversarial network models in deep learning, mainly covering the problems solved by the inpainting methods, the advantages and disadvantages of different inpainting methods, and as well the methods to be improved, etc. Finally, we analyze the difficulties in image inpainting, and point out the development in the future.