Video Saliency Map Detection by Dominant Camera Motion Removal

  • Author(s):

Chun-Rong Huang, Yen-Yu Lin, Zhi-Xiang Yang, and Yun-Jung Chang


    THE human visual system (HVS) perceives the world and provides visual information for human beings. It is known that only a small fraction of the observable area is critical for humans to understand and interpret the world. Hence, recognizing saliency maps, which record the distribution of human’s attention, is helpful in understanding how HVS works and analyzing the content of images/videos. We present a trajectory based approach to detecting saliency regions in videos by dominant camera motion removal. Our approach is designed in a general way so that it can be applied to videos taken by either stationary or moving cameras without any prior information. Moreover, multiple saliency regions of different temporal lengths can also be detected. To this end, we extract a set of spatially and temporally coherent trajectories of keypoints in a video. Then, velocity and acceleration entropies are proposed to represent the trajectories. Specifically, one-class SVM is employed to remove the consistent trajectories in motion. It follows that the saliency regions could be highlighted by applying a diffusion process to the remaining trajectories. The annotated videos are then used for performance evaluation and comparison. The promising results on various types of videos demonstrate the effectiveness and great applicability of our approach. Here is our dataset.


  • If you use the dataset, please cite our papers in the publication section.

  • The windows GUI demo program is available here. The program has been modified to obtain explicit boundaries of salient objects for comparison and achieves better performance compared to the original version.

Demo Videos


[1] Chun-Rong Huang, Yun-Jung Chang, Zhi-Xiang Yang, and Yen-Yu Lin, "Video Saliency Map Detection by Dominant Camera Motion Removal, " To Appear in IEEE Transactions on Circuits And Systems For Video Technology.