Research On Saliency Detection Based On Propagation Mechanism

Image has become essential communication medium in people's life.However,the increases of image data and the improvement of resolution ratio have put forward higher requirements for software algorithms and hardware technology.Saliency detection aims to extract the most interested areas or objects of the human eye,which undoubtedly provides great convenience for subsequent processing.In recent years,saliency propagation mechanisms have achieved good results,however,it exists numerous limitations.Aiming at the shortcomings of existing work,this paper proposes two saliency propagation mechanism algorithms.(1)This paper proposes a saliency propagation algorithm based on local structure,which is utilized to address problem that existing works hardly suppress background noises.First,two prior acknowledges are integrated into coarse saliency map,even including proposed spatial distribution saliency map based on central compactness rule.Then,this paper proposes a local structure propagation mechanism to optimize coarse map,the performance of obtained saliency map is further enhanced.(2)This paper proposes a saliency propagation mechanism based on effective and contrast,which is utilized to address problem that existing work hardly stay salient object's structure.This paper firstly provide a method eliminated foreground noises of image boundary to construct more effective background-based map,which is integrated Objectness map into coarse saliency map.Then,this paper proposes a saliency propagation mechanism based on effective local and contrast information to optimize coarse saliency map.Finally,an integration mechanism is proposed to further enhance performance.Two proposed methods are compared with other 11 saliency detection methods in five datasets,experimental results demonstrate the superiority of proposed algorithms.Meanwhile,parameters and each component about proposed algorithms are also deeply analyzed.In the end,proposed propagation mechanism is applied to existing methods,experimental results indicates that proposed algorithms has classic adaptivity.