Saliency Detection via Background Seeds Combined with Bayesian Model
Linhua Jiang, Pengfei Bai, Xiao Lin · 2016
To detect the salient objects of an image and emphasize the contrast between the salient objects and background regions, we propose a new method named Saliency detection via background seeds combined with Bayesian model. First, we exploit the simple linear iterative clustering (SLIC) to segment the image into superpixels. Then we choose the superpixels on the borders of an image as background seeds to calculate the prior map based on background. Third, we obtain the pseudo-ground-truth map by setting a threshold to segment the prior map which is the rough position of the salient region. Next, we compute the likelihood probability of the salient region. Finally, we use the prior map and likelihood probability to compute the final saliency map based on the Bayesian model. Experimental results on several available datasets demonstrate that the proposed algorithm achieves favorable results in term of the precision and recall.