BINet: A binary inpainting network for deep patch-based image compression

Recent deep learning models outperform standard lossy image compression codecs. However, applying these on a patch-by-patch basis requires that each patch be encoded and decoded independently. The influence from adjacent patches is therefore lost, leading to block artefacts at low bitrates. We propose the Binary Inpainting Network (BINet), an autoencoder framework which incorporates binary inpainting reinstate interdependencies between patches, for improved patch-based of still images. When decoding patch, BINet additionally uses binarised encodings surrounding guide its reconstruction. In contrast sequential methods where are based previons reconstructions, operates directly codes without access original or reconstructed data. Encoding can performed in parallel. demonstrate improves quality competitive codec across range levels.