Intelligent image inpainting based on a brain-mimicking Recognition-Mining-Synthesis network

In this paper, a new image inpainting method based on a brain-mimicking Bayesian network is proposed. It features the hierarchical-spatial-temporal prediction for simulating brain´s neural projections. The trained network can recognize image and automatically mine for defects by its attention model, before synthesizing lost image parts. Difficult inpainting problems, like images with important features lost, are shown to have good results.