GAN-based Image Inpainting Techniques: A Survey
Aishwarya Mohod, Piyoosh Purushothaman Nair · 2024
Image inpainting is the process of completing or reconstructing areas in an image that are absent or damaged using information extracted from the surrounding parts of the same image. The output inpainted images should be realistic and challenging to discern from the original images. Image degradation can occur due to various factors, posing a significant challenge for image inpainting, which makes it a vital and intricate task in the domain of image processing. This paper centers around a variety of image inpainting techniques that rely on Generative Adversarial Networks (GANs), which have gained prominence lately. The paper further discusses various datasets available for image inpainting. An overview of commonly used evaluation metrics is provided, and the performance of various GAN-based image inpainting methods in terms of these evaluation metrics is discussed.