Smoothness加强的全局和局部显著目标检测

摘要

显著目标检测是计算机视觉的重要组成部分。针对基于对比度的方法存在的前景和背景容易被误检的问题，提出Smoothness 加强的全局和局部显著目标检测的方法。在全局对比度检测过程中引入中心先验知识，在局部对比度检测过程中引入Compactness 特征，再使用 Smoothness 特征加强全局的显著性及局部的显著性，最后将全局显著图和局部显著图进行线性融合。在MSRA-1000、ECSSD 数据集上的评估中，该算法有更高的准确率，在 CSSD 数据集上能和最先进算法相媲美。实验表明，从全局和局部两个角度出发的显著性检测的方法能够有效的互补，Smoothness 能够有效加强前景和背景的差异性，并且有效纠正一些误检现象，从而取得更好的结果。%Salient object detection is an important topic in computer vision.In order to solve the problem of detection errors which traditional contrast-based methods have in detecting foreground and background,we proposed a new method of global and local salient object detection with smoothness strengthened.First,it introduces centre prior knowledge to the process of global contrast detection and introduces compactness feature in the process of local contrast detection.Then it uses smoothness feature to strengthen the salience of global and local detection respectively.Finally,it carries out linear fusion of global salient graph and local salient graph.In comparison assessments on MSRA-1000 and ECSSD datasets,the proposed algorithm achieved higher precision both.And on CSSD dataset,its performance was comparable to the state-of-the-arts approaches.Experimental results demonstrated that the salience detection from two perspectives of global and local can effectively complement to each other.Smoothness feature can effectively strengthen the difference of foreground and background,and corrects some detection errors as well,so that achieves better results.