Improved coordinate-based image and video cloning algorithm

Abstract

Seamless cloning algorithm for image and video is an important editing operation. Recently [Farbman et al. 2009] introduced an alternative which exploits the Mean-Value Coordinates (MVC) rather than solving a large linear system to perform a weighted sum of values along the boundary. It demonstrated result as good as that of the earlier techniques in a remarkably short time and ease of implementation. Seamless video object cloning was also discussed in [Farbman et al. 2009], where they used a simple exponential decay of the previous frames. In this paper, we propose an improved MVC algorithm in the two aspects. First, we provide an alternative sampling algorithm which make it possible to handle concave regions that were problematic in [Farbman et al. 2009]. And we propose more effective temporal smoothing method for video cloning than a simple exponential decay method. Our method is not only fast enough to carry out interactive video cloning, but also it can reduce temporally noticeable seams of video.