Saliency Optimization Integrated Robust Background Detection with Global Ranking

Image saliency detection plays an important role in the field of computer vision. In order to make the saliency detection achieve better results, this paper proposes an algorithm that combines the global cue with the robust background prior measurement. Specifically, we first get a global ranking of superpixels by absorbing time in Markov chain, which is the absorbing nodes represent the superpixels of the fictitious boundary, and the transient nodes denote the others. Then, the global similarity of transient node can be measured by its absorbing time, so a global ranking for each transient node can be calculated, which is called as global cue in this paper. Finally, considering the remarkable energy optimization model, we integrate the robust background prior measurement with calculated global cue to form a new optimization model for saliency detection. In conclusion, our method is better than some typical significant detection algorithms on several datasets through the experimental verification.