Saliency Optimization Based On Background Detection

All the time,content-based image retrieval(CBIR)is a hot research area,which was used widely in various fields.Recent years,the boundary prior,or background information have exploited in salient object detection progresses.The paper extracts the target region from the background.The background is identified first,and the remainder is considered the target region.It is the focus of this paper that the computer how to extract the salient region of the image quickly and accurately.So the following work is done by the paper.Firstly,based on the existing boundary prior and background information,we changing the constraints of background prior,ie,an image patch is background only when the region it belong to is heavily connected to the image boundary,and an image patch that is slightly connected to the boundary is most likely the target.And we regard it as a new measure.Instead of assuming the image boundary is background,or an image patch is background if it can easily be connected to the image boundary.Compared to existing methods which exploit boundary prior having a higher robust.Secondly,in this paper,we use impact factor formula that is obtained from progresses of calculating to update synchronously in the saliency map that is generated from previous step.The method of this,make the correlation between adjacent pixels to strengthen,then,we can obtain a more highlight object in saliency map and more clearly at the edges.Through experiments,it is proved that the proposed algorithm is very good to solve the defect that the saliency of the original algorithm is not clear.Finally,According to the Bayesian formula,we introduce a saliency integration mechanism based on Bayesian fusion to combine the saliency map that generated from update synchronously and the saliency map that obtained from the algorithm of Saliency Optimization from Robust Background Detection.At last,we obtained the saliency map with clear contour and the interior is more homogeneous and compact.Compared with other algorithms,the final result has certain superiority on the same dataset.