语义信息引导下的显著目标检测算法  

Salient Object Detection Algorithm Under Guidance of Semantic Information

作　　者：肖锋[1] 李茹娜 XIAO Feng;LI Runa(School of Computer Science and Engineering,Xi’an Technological University,Xi’an 710021,China)

机构地区：[1]西安工业大学计算机科学与工程学院,西安710021

出　　处：《计算机工程》2019年第4期248-253,共6页Computer Engineering

基　　金：国家自然科学基金(61572392;61671362);陕西省自然科学基础研究面上项目(2017JC2-08)

摘　　要：针对现有显著性检测方法在凸显目标完整性和背景噪声抑制方面性能较差的问题,提出一种显著目标检测算法。构建改进的全卷积神经网络,捕获图像中的语义信息,生成高层语义初步显著图。利用语义知识引导流形排序进行优化,实现显著目标的边缘传播。融合不同尺度下的显著图,完成显著目标检测。在ECSSD、DUT-OMRON、SED2数据集上进行实验,结果表明,与最大对称环绕、主成分分析等算法相比,该算法检测出的显著目标更加完整,在复杂场景下检测结果鲁棒性更好。In view of the poor performance of the existing saliency detection method in object highlighting and background noise suppression,a salient object detection algorithm is proposed.An improved Fully Convolutional Neural Network(FCNN) is constructed to capture the semantic information in the image,which can generate the high-level semantic preliminary saliency maps.The semantic knowledge is used to optimize the Manifold Ranking (MR),and then achieve the edge propagation of salient object.The saliency maps under different scales are fused to realize salient object detection.Experiments are carried out on ECSSD,DUT-OMRON and SED2 datasets,the results show that compared with algorithms such as Maximum Symmetric Surround(MSS)and Principal Component Analysis(PCA),the salient objects detected by the algorithm in this paper are more complete and the detection results are more robust in complex scenes.

关 键 词：显著目标检测 语义信息 流形排序 全卷积神经网络 目标感知 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]