Advanced image completion techniques based on deep learning

Deep artificial neural networks have been successfully used in many different problems in recent years. Convolutional Neural Networks (CNN), which stand out with their success in image classification problems, and Generative Adversarial Networks (GAN), which are used to produce realistic images, can be given as examples. Image completion is another area where deep learning is used and successful results are obtained. Specifically, in the outpainting problem, it is a very challenging problem to complete the image in a meaningful and coherent way because the filled part is large and comes to the outer part of the image. In this study, image outpainting techniques are implemented with deep learning. Image completion was performed using autoencoders and enhancements were made using discriminative networks. In addition, an edge-completing deep learning model was developed uniquely for this study and the edge map obtained was used in the outpainting process.