A novel video saliency map detection model in compressed domain

Abstract

Saliency detection in videos has attracted great attention in recent years due to its wide range of applications. In this paper, a novel spatiotemporal saliency detection model based on clustering is proposed. The discrete cosine transform coefficients are used as features to generate the spatial saliency map firstly. We utilize 2D Gaussian function to estimate the absolute feature difference in consideration of video resolution. Multiple spatial saliency maps which indicate different features are constructed and linearly combined to obtain the overall spatial saliency map. Then, a hierarchical structure is utilized to obtain the temporal saliency map using the extracted motion vectors that belong to the foreground. In addition, spatial and temporal saliency maps are clustered into non-overlapping regions automatically based on the histogram of each saliency map. Finally, an adaptive fusion method is used to merge clustered spatial and temporal saliency maps of each frame into its spatiotemporal saliency map. Based on the experimental results obtained in our study, the performance of the proposed approach is better than those of the other compared approaches.