DIUN:Deeper Inception U-Network for Recovering Partial Pixelated Images

In our daily life,it is nothing strange to see pixelated images that are spoiled artificially to hide certain information for protecting privacy or pixelated deliberately to cover up bad behaviors even crimes.To prevent these phenomena and recover the true information from pixelated images,it is meaningful to research an effective reconstruction method for recovering pixelated images.This paper aims at recovering the artificial partial pixelated images via deep learning(DL).To abstract more abundant features and enhance the repair ability of DL model,we propose a new DL structure,called deeper inception U-Net,to act as the generator of a generative adversarial network.We combine the feature loss with structural similarity index measure loss as the context loss to minimize the distance between feature maps of clear images and the generated images,which helps to improve the quality of repair images.After obtaining inception features,we use fusion layer to adaptively learn features in each inception block.To evaluate the performance of our model,we introduce a new home dataset that contains 10174 clear home images with corresponding pixelated images.A series of experiments show that our model has ability to rebuild pixelated images.