A Saliency Map Fusion Method Based on Weighted DS Evidence Theory

In this paper, we propose a weighted Dempster-Shafer (DS) evidence theory-based fusion algorithm to take advantages of state-of-the-art salient object detection methods. First, we define the mass function value for each saliency detection method to be fused at the pixel level, based on which we further calculate the similarity coefficient and similarity matrix. The credibility of each saliency detection method will be computed by considering to what degree it is supported by other saliency detection methods. Second, using the credibility of each image saliency detection method as the weight, we compute the weighted mass function value of each method and get a saliency map. Third, we use the synthetic rules of DS evidence theory to fuse the weighted mass function values and get the other saliency map. The final saliency map will be obtained by fusing the aforementioned two saliency maps. Extensive experiments on three publicly available benchmark datasets demonstrate the superiority of the proposed weighted DS evidence theory-based fusion model against each individual saliency detection algorithm in terms of three evaluation metrics of precision-recall rate, F-measure, and average absolute error. The saliency map after fusion utilizing weighted DS evidence theory is closer to the ground-truth map.