Content-aware copying and pasting in images

We present a content-aware image copy-and-paste technique which combines ideas from both matting and gradient-based methods. We modify the diffusion process used in the gradient-based approach to use the alpha matte for the cloned area as a weight function to control intensity interpolation. This ensures that the color style of the significant parts of the selected region is preserved when pasting. We use a framework based on mean-value coordinates to implement our approach, allowing us to provide a parallel implementation suitable for use on a GPU. Experimental results show the advantages of our method.