基于新图模型的显著性检测  

Significance Detection based on New Graph Model

作　　者：陈照兴 张巧荣[1] Chen Zhaoxing;Zhang Qiaorong(Henan University of Economics and Law,Zhengzhou Henan 450000,China)

机构地区：[1]河南财经政法大学,河南郑州450000

出　　处：《信息与电脑》2017年第23期53-56,共4页Information & Computer

基　　金：河南省高等学校重点科研项目计划项目(项目编号:18A520016)

摘　　要：利用新的图形结构和背景先验提出了一个无监督的自底向上的显著性检测方法。输入图像被表示为一个用超像素块作为节点的无向图。将每个节点覆盖区域的颜色、对比度和纹理信息作为特征向量,结合一种新的图形模型,有效地提出了捕捉局部和全局显著性的线索。为了获得更准确的显著性估计,使用强大的背景测量来优化显著图。在基准数据集上的综合评价表明,本算法优于大部分传统的算法。An unsupervised bottom-up saliency detection method is proposed by using the new graph structure and background.The input image is represented as an undirected graph with superpixel blocks as nodes.The color,contrast and texture information of each node area are taken as the feature vectors,and a new graphical model is proposed to effectively capture the clues of local and global saliency.In order to obtain more accurate saliency estimation,strong background measurements are used to optimize saliency maps.The comprehensive evaluation on benchmark data sets shows that the proposed algorithm is superior to most traditional algorithms.

关 键 词：图像显著性 新图模型 背景先验 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]