Learning video saliency from human gaze using candidate selection