Robust video fingerprinting based on visual attention regions