Fast Calculation Method of Video Saliency Based on Temporal and Spatial Edge-Preserving Filter

Abstract

How to detect video saliency accurately is an very challenging problem in computer vision, and has gained great research and application values. A fast video saliency detection method, which based on spatio-temporal edge-preserving filter, was proposed in this paper. Firstly, object appearance features and motion features were extracted in video frames. Secondly, a new spatio-temporal edge-preserving filter was constructed by combining object temporal information in video. Then, the video saliency detection is formulated as an energy minimization problem, which fuses the appearance and motion features, and stand out salient object in video uniformly by using the spatio-temporal edge-preserving filter. Finally, extensive experiments, including comparisons analysis with three video saliency detection approaches and components analysis of the proposed approach, on three datasets suggest that the proposed approach nearly outperforms other approaches in both accuracy and efficiency.