Three-dimensional movies, television programs, and games are becoming increasingly popular. This trend fosters an urgent need for tools to edit stereoscopic 3D images. “Copy & paste” is one of the most-wanted editing functions. However, 2D copy & paste cannot be extended to 3D in a straightforward manner, since a 3D operation has to deal with depth. In this paper, the authors present a stereo billboard technique to perform 3D copy & paste.
Three-dimensional copy & paste poses at least three major challenges. First, the disparity of the copied object has to be consistent with the target region (for example, the shape and the stereo volume should be preserved.) Second, occlusion should be correctly handled because it is an important depth cue for humans and humans are very sensitive to errors in it. Finally, the depth information that one can reconstruct from stereo images is often inaccurate and noisy due to the image noise and the lack of corresponding points in stereo images. A good 3D copy & paste technique must be insensitive to these imperfections. This paper proposes a technique that addresses all three challenges.
The proposed technique first leverages user interaction and a previously proposed mean-shift segmentation method to select an object to copy. Users can then rotate the copied object under certain constraints and paste it on the target region. The proposed billboard technique ensures that the volume of the object won’t change during this process. The objects are sorted according to depth so that occlusion can be rendered correctly.
This technique is seemingly designed to be an essential tool for 3D image editing. It may not work when one wants to manipulate the surfaces of complicated shapes or rotate a selected object by a large angle. However, these limitations could be addressed in future work or other 3D editing tools.