Salient map extraction based on motion history map

Motion saliency is a key component for video saliency model, and attracts a lot of research interest. However, the existing methods will lose the foreground objects or parts of the foreground objects when they stop moving for a certain period of time, and the interior of objects will be marked un-salient unless the moving objects are sufficiently textured. In this paper, a novel saliency model based on motion history map is proposed. We generate a spatial saliency sub-map by pixel-wise self-information and global contrast, and use it to adaptively adjust the effect period of motion history map, which makes the foreground objects are still marked silent for a long time after they stop moving, while the background noise is soon marked un-silent. Otherwise, a fast region growing method based on the spatial saliency sub-map is applied to mark the interior of moving objects salient. The experiment results show that our proposal can achieve better saliency map extraction performance.