Bottom-Up Saliency Detection Based On Background Background Prior And Object Information

The real world is full of a large amount of information,and most of the human perception of external information is based on human visual processing,which is mainly due to the human visual system(HVS)having a strong ability of information processing and perception.For a given scene,human attention tends to focus on some of the most important objects known as salient objects,thus automatically ignoring a large amount of insignificant information.In scenarios,the objects that attract human attention are often referred to salient objects.In the face of the exponential growth of images every day,scientists have conducted extensive research on image/video significance detection by simulating the human visual mechanism,trying to guide the computer to solve visual tasks quickly and accurately.At present,as a pre-processing stage,saliency detection has been widely applied to other visual fields,such as image/video segmentation,object tracking,object recognition,content-aware image cropping,and soon.Although many excellent saliency detection models have been developed,it is still a challenge to design algorithms that can efficiently and accurately detect salient objects in scenes due to the complexity of scenes and the diversity of objects.This paper mainly focuses on the research of salient object detection.By analyzing the shortcomings of existing models,four salient object detection algorithms are proposed,which can not only accurately highlight the salient object in the scene,but also effectively suppress background noise.The main contributions of this paper are as follows:(1)A salient object detection algorithm based on background template and energy function is proposed.The algorithm defines the background template to calculate the background saliency,which effectively alleviates the interference caused by the foreground and boundary contact.At the same time,we propose a method based on size control to obtain the optimal threshold and acquire the foreground seeds by binarization.Finally,a minimum energy function is designed to further improve the accuracy of saliency detection.The effect of the proposed method is better than the current unsupervised salient object detection algorithm.(2)A salient object detection algorithm based on multi-scale fusion background prior and label propagation is proposed.In order to extract reliable background information,a robust background seed selection mechanism based on convex hull and corner information is proposed for the first time.In addition,the combination of high-level feature of Objectness and low-level features is used for saliency detection,which effectively compensates for the defects of low-level features.In order to overcome the shortcomings of single-scale detection,a novel fusion algorithm was designed to deal with the results of different scales to get the final result.(3)An unsupervised salient object detection algorithm based on aggregated multi-level cues is proposed.Object proposals have high-level features that facilitate saliency detection,but too many object proposals are not only mixed with useless noise information that is detrimental to detection,but also the excessive number will seriously reduce the speed of detection.To this end,this paper proposes a robust proposal ranking scheme by considering various spatial saliency clues at the object level.In order to improve the performance of the algorithm,in addition to object-level clues,the method also integrates super-pixel and pixel-level clues.Comparative experiments verify the effectiveness of the model.(4)A salient object detection algorithm based on background divergence and foreground compactness is proposed.Considering the complementary characteristics of semantic information and low-level information,this paper calculates regional feature similarity by integrating semantic information and low-level information.From the perspective of global image,robust background seeds are extracted by using background divergence,and compact and coherent foreground regions are generated by considering the spatial compactness and rarity of salient objects.In addition,this paper also designs a versatile propagation algorithm to improve the accuracy of saliency detection.A large number of experimental results show that the proposed algorithm has better performance than existing algorithms.