A Framework for Interactive Realtime Image Editing

Abstract

We present a complete framework for interactive image editing with a focus on quickly obtaining results based on user input via brushstrokes. The goal is to show feedback to changing input in realtime while maintaining high quality output. We achieve this by extracting vital information in the form of local color clusters and a corresponding locally linear structure from the input image. With this information the problem of propagating any sort of features provided by user input can be executed conveniently and very efficiently. One of the major merits of our approach is that the locally linear structure does not impose any restrictions on the cluster structure. The fully automatically created structure essentially allows for a flexible scaling of the overall computational cost for propagating user edits. If needed, a more detailed output can be obtained by enforcing the creation of smaller clusters.