Image Outpainting using Wasserstein Generative Adversarial Network with Gradient Penalty

With advancements in AI technology, machines can perform or even mimic tasks that humans can do. One of its achievements can be seen in image generation, one of it being Image Inpainting (completion). In Image Inpainting, AI is used to complete missing data in an image. This is an extensive field of research, but its contemporary field, i.e. image outpainting, is not a well-researched one. In Image Outpainting (extrapolation) the image is extended beyond its borders. This is similar to our brain picturing the whole image of an object that is partially seen through a gap. This task can be achieved by using Generative Adversarial Networks (GANs). Compared to Inpainting, the biggest challenge is to achieve spatial correlation between the generated image and the ground truth image. Also, the process of overcoming this challenge is also sometimes affected because of the training instability of GAN. With the help of Wasserstein GAN (WGAN), the above issue can be solved. So, a model is proposed based on the Wasserstein GAN with Gradient Penalty (WGAN-GP) algorithm and deep convolutional neural networks for image outpainting using a dataset on natural images. From this proposed model it is found that the results of WGAN-GP algorithm was better than GAN algorithm in various aspects.