Video Object Segmentation for Content-Aware Video Compression

We present a method for extracting foreground objects from video and its application to content-aware video compression. Our method uses trimaps inferred from background subtraction to represent the foreground-background relationship. The appearance of foreground and background are modeled with Radial Basis Functions initialized from the background substraction step. Finally, Graph Cuts are used to compute a binary mask. Our method is fully automatic, fast, and does not make restrictive assumptions about object motions. In experiments on standard data sets, the proposed approach achieves comparable results to state-of-the-art video object segmentation methods but our method is much faster. We also demonstrate an application of the proposed method to content-aware video compression.