Research On Image Saliency Detection Technology Based On Robust Front Background

With the emergence of a large amount of multimedia data such as pictures,image saliency detection which is intended to extract an attractive area and filter redundant information is playing an increasingly important role.As one of the preprocessing work in the field of computer vision,various image saliency detection algorithms and related applications have been proposed.Aiming at the problem that objects are not salient enough and the background suppression is insufficient or wrong for some traditional saliency detection algorithms,this paper respectively proposes saliency detection algorithms based on robust foreground,robust background and combination of foreground and background.1.Image saliency detection algorithm based on robust foreground.The existing foreground-based Bayesian saliency detection algorithm is easy to mistakenly select the background area in the convex hull as a foreground when the background is complicated,resulting in a large deviation of the saliency detection result.Aiming at this problem,this paper proposes a method for extracting the foreground in convex hull based on boundary connectivity.This method mainly calculates the degree of connection between a region and the image boundary,chooses the regions with the smaller connectivity as a foreground in the convex hull,and then adopts Bayesian model to achieve accurate detection of salient object and suppression of background.2.Image saliency detection algorithm based on robust background.The traditional background-based manifold ranking saliency detection algorithm uses the image boundary as the background seeds,but the background seeds are not accurate enough when the target touches the image boundary,which leads to the object not being salient.In order to overcome this problem,this paper proposes a method to extract robust background based on color difference between background and foreground.This method mainly clusters image boundaries and removes the classes with a small distance from the robust foreground,thus improving the performance of the manifold ranking saliency detection algorithm.3.Unlike most of current saliency detection algorithms,which only utilize a single foreground or background prior for saliency detection,this paper proposes a saliency detection algorithm based on the combination of foreground and background to makefull use of the complementarity of these two priors.The method firstly uses the foreground for manifold ranking saliency detection,and then fuses the obtained saliency map with the background-based saliency map to obtain a more stable saliency detection result.In this paper,the effectiveness of the saliency detection algorithms based on robust foreground and robust background are verified by experiments on the public dataset.At the same time,the saliency detection algorithm based on the combination of foreground and background is compared with some classic algorithms,which shows good detection performance.