Image inpainting using artificial intelligence

The goal of the image inpainting process is to reconstruct images in such a way that the completed area is undetectable by the observer. Recently, with the development of image processing techniques, image reconstruction methods have become an important and challenging research topic. In this paper, the topic of image inpainting using artificial intelligence is introduced. State-of-the-art image reconstruction methods were analyzed and compared with each other, and then one of them was chosen as the base method for the experiments presented in this paper. In the practical part, the influence of the perceptual loss function on the missing data completion process was investigated. The described methods from the deep learning-based group are Context Encoders and Globally and Locally Consistent Image Completion. Those belonging to the attention-based methods are Shift-Net, Contextual Attention, and Coherent Semantic Attention. Finally, all methods were compared with each other and, as they gave the best results, Coherent Semantic Attention was chosen as the base method for further research. The experiments performed and their qualitative and quantitative results are described. Quality metrics, previously described, were used to evaluate the correctness of each experiment.