摘要
由于视觉注意预测能够快速、准确地定位图像中的显著区域,因此将视觉注意中的频域信息融入显著性目标检测中,从而有效地在复杂场景中检测显著性目标。首先,采用改进的频域检测方法对图像进行视觉注意预测,将该频域信息融入Focusness特征中计算得到频域信息聚焦特征,并将此特征与颜色特征进行融合得到前景显著图。然后,对RBD背景进行优化,得到背景显著图。最后,对前景显著图、背景显著图进行融合。在ESSCD,DUT-OMON两个具有挑战性的数据集上进行了大量实验,并采用PR_Curve,F-Measure,MAE对结果进行了评估,结果表明,所提出的方法要优于6种对比方法(HFT,PQFT,HDCT,UFO,DSR和RBD),并且能够处理复杂场景的图像。
Since visual attention prediction can locate the salient area of image quickly and accurately,in this paper,the frequency domain information of visual attention was integrated into the saliency object detection,to detect the saliency object effectively in the complex scene.Firstly,the improved frequency domain detection method is used to predict the visual attention of image,and the frequency domain information is blended into Focusness feature to calculate the frequency domain information focusness feature,which is combined with the color feature to generate the foreground sa-liency map.Next,the RBD background is optimized to generate the background saliency map.Finally,the foreground saliency map and background sa-liency map are fused to generate saliency map.A large number of experiments were carried out on two challenging datasets(ESSCD and DUT-OMON),and the results were evaluated by PR curve,F-Measure and MAE.Experimental results show that the proposed method is better than HFT,PQFT,HDCT,UFO,DSR and RBD,and it can deal with the images with complex scenes.
作者
袁小艳
王安志
王明辉
YUAN Xiao-yan;WANG An-zhi;WANG Ming-hui(School of Intelligent Manufacturing,Sichuan University of Arts and Science,Dazhou,Sichuan 635000,China;School of Computer Science and Technology,Southwest Minzu University,Chengdu 610041,China;College of Computer Science,Sichuan University,Chengdu 610064,China)
出处
《计算机科学》
CSCD
北大核心
2018年第10期261-266,共6页
Computer Science
基金
国家重点研究与发展计划(2016YFB0700802
2016YFB0800600)
国家海洋局海洋遥感工程技术研究中心创新青年项目(2015001)
四川省教育厅一般项目(18ZB0509)
四川文理学院智能制造产业技术开发研究专项项目(2017ZZ006Y)资助
