A video saliency detection method based on spatial and motion information

Saliency detection for images and videos has become increasingly popular due to its wide applicability. In this paper, we propose a novel bottom-up video saliency extraction method, which includes two parts: static saliency detection and dynamic saliency detection. Firstly, we consider the static saliency detection as a classification problem: a scene can be divided to saliency region and non-saliency region. To solve such problem, we use a frequency-based sampling method and then introduce a Canonical Correlation Analysis (CCA)-based classification strategy using multiple cues. Secondly, we combine motion feature with the spatial feature to represent dynamic saliency in temporal domain. We test our method in both image based saliency extraction and video saliency extraction. The results illustrate its high performance in both two tasks.