Salient Region Detection in Video Using Spatiotemporal Visual Attention Model

Abstract Salient region detection is very useful in video analysis. A salient region detection method based on spatiotemporal visual attention model is proposed in this paper. Visual attention mechanism is used to generate saliency map of the image sequence. Spatial saliency map is computed in accordance with some predefined features including intensity, color and orientation. Temporal visual saliency of each pixel is calculated using motion feature. Both spatial saliency map and temporal saliency map are combined to obtain the overall visual saliency map. Then salient regions are detected based on the visual saliency map. The proposed salient region detection method has been tested on many videos. The experiment results show that the presented method is valid and effective. The detection results are consistent with human visual system.