Saliency Detection Based On Convex Hull Background Prior And Target Prior

Image saliency detection has become one of the most important research topics in recent years.The purpose is to quickly obtain useful information in images.Aiming at the disadvantage of the current saliency detection method that the target area is not bright,this paper proposes a saliency detection method based on the convex hull background priori and the target priori.The main process is to construct the background priori map and the target priori map separately,and adopt Bayesian model to fuse them.In order to obtain accurate and reliable saliency maps,the background template information and target template information are fully obtained.1.Saliency detection based on background area outside convex hull.Compared with the traditional method of using image surroundings as background template,this section proposes a method of using super-pixels outside convex hull as background template to construct background priori map.This method greatly improves the disadvantage of using image surroundings as background template.Firstly,the salient points of the image are acquired,the salient points of the edge are eliminated and the convex hull is obtained.Secondly,the image is segmented into super-pixels,and the super-pixels outside the convex hull are used as background template.Then the whole image is projected to the background template to calculate the reconstruction error and obtain the background priori image.2.Saliency detection based on target region inside convex hull.In this section,we propose a method to construct a priori image by using the super-pixels inside convex hull as the target template.The main purpose of this method is to project the whole image onto the target template and calculate the reconstruction errors of the target template and the whole image,so as to obtain the reconstruction error map.Then,a Gaussian filtering model based on target bias in convex hull is proposed,and the final target priori graph is obtained by combining the anti-reconstruction error graph.3.Current optimization methods have the problem of using single background or target.Due to the respective advantages of background and target priori maps,this section proposes a method of fusing background priori maps with target priori maps and the final saliency map is obtained by using Bayesian model.The algorithm in this paper has carried out a large number of experimental evaluation on four open datasets,and has carried out phased experimental verificationon the process of constructing background priori map and target priori map respectively.At the same time,the final experimental results are compared with the existing classical algorithms.Compared with other algorithms,the proposed algorithm shows good detection performance.