Automatic video segmentation plays an important role in a wide range ofcomputer vision and image processing applications. Recently, various methodshave been proposed for this purpose. The problem is that most of these methodsare far from real-time processing even for low-resolution videos due to thecomplex procedures. To this end, we propose a new and quite fast method forautomatic video segmentation with the help of 1) efficient optimization ofMarkov random fields with polynomial time of number of pixels by introducinggraph cuts, 2) automatic, computationally efficient but stable derivation ofsegmentation priors using visual saliency and sequential update mechanism, and3) an implementation strategy in the principle of stream processing withgraphics processor units (GPUs). Test results indicates that our methodextracts appropriate regions from videos as precisely as and much faster thanprevious semi-automatic methods even though any supervisions have not beenincorporated.
展开▼