Improved Saliency Detection based on Bayesian Framework for Object Proposal

In this paper, a new method is proposed for object proposal based on saliency detection. First, a novel method is proposed to measure the global spatial compact distribution of the color components in an image. The saliency detection method proposed on the basis of Bayesian improves the estimation of prior probability and likelihood of observations by means of an optimized boundary connectivity measure. Second, based on the saliency map of the method proposed, the object proposal is given with the bounding box, through non-maxima suppression sampling strategy. Both, the saliency detection method and the object proposal method, are evaluated and compared with state-of-the-art results on standard databases. The experimental results on the challenging PASCAL VOC2007 data set show that the detection rate of the object proposal method proposed can reach 93.4% for the first 1000 windows proposed.