Cross Scene Video Foreground Segmentation Via Co-Occurrence Probability Oriented Supervised and Unsupervised Model Interaction.