Spectral Video Matting
Martin Eisemann, J. M. Wolf, Marcus A. Magnor · 2009
We present a new, simple-to-use and rapid approach to video matting, the process of pulling a highquality alpha matte from a video sequence. Our approach builds upon techniques in natural image matting, namely spectral matting, and optical flow computation. No additional hardware, despite a single camera, is needed, and only very few and intuitive user interactions are required for foreground estimation. For certain scenes the approach is able to estimate the alpha matte for a video, consisting of up to 102 frames, without any user interaction at all.