Image Inpainting Based on Interactive Separation Network and Progressive Reconstruction Algorithm

Recently, learning-based image inpainting has gained much attention. It widely utilizes an auto-encoder structure and can obtain compact feature representation in the encoder to achieve high-quality inpainting. Although this approach achieved encouraging results, it inevitably reduces high-resolution due interval downsampling. In order solve problem excellent effect, paper proposes a brand-new generative network, Interactive Separation Network, which retains information extracts semantic features from corrupted images. Furthermore, also discusses network designs with different complexity application scenarios. Finally, improve effectiveness robustness of our proposal large regions image, we further propose flexible highly reusable reconstruction scheme complete prediction process gradually. Experiments show that proposed generation significantly quality repaired The method outperforms state-of-the-art approaches quality.