电子测量与仪器学报

徐涛,马玉琨.基于引导传播和流形排序的协同显著性检测方法[J].电子测量与仪器学报,2017,31(12):1999-2008

基于引导传播和流形排序的协同显著性检测方法

Co-saliency detection algorithm based on bootstrap propagation and manifold ranking

DOI：10.13382/j.jemi.2017.12.017

英文关键词:visual saliency co-saliency detection bootstrap propagation manifold ranking Bayesian inference

基金项目:国家自然科学基金（61304061）、河南科技学院“标志性创新工程”、2016年河南省产学研合作项目（162107000058）资助

作者	单位
徐涛	河南科技学院机电学院新乡453003
马玉琨	2. 北京工业大学信息学部北京100124；3. 河南科技学院信息工程学院新乡453003

Author	Institution
Xu Tao	School of Mechanical and Electrical Engineering, Henan Institute of Science and Technology, Xinxiang 453003, China
Ma Yukun	2. Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China; 3. School of Information Engineering, Henan Institute of Science and Technology, Xinxiang 453003, China

摘要点击次数: 2489

全文下载次数: 6991

中文摘要:

提出一种基于图像间显著性引导传播和图像内流形排序的两阶段引导方法，充分挖掘单幅图像显著性引导传播机理，提高面向群组图像协同显著性检测算法的精确度和实时性。对N张群组图像中的任意一幅图像，第一价段借助单幅图像显著性探索其与组内其他图像两两间的共同相似性属性，获取N-1张初始协同显著性图。为了有效抑制非相似区域的背景干扰，在第二阶段中，通过流形排序（EMR）算法，计算N-1张前景显著性图每个像素点的排序值，以更新之前的显著性检测结果，恢复出第一阶段中误检为背景的相似性区域。最后在基于贝叶斯理论的融合算法框架下实现最终协同显著性图的获取。基于iCoseg和MSRC数据库进行评测，所提算法在综合指标F值和受试者工作特征（ROC）曲线下面积（AUC）等评价指标方面一致优于现有5种协同显著性检测算法。基于真实场景的实验结果从普遍适用性角度对本文算法做了进一步验证。

英文摘要:

A two stage guided co saliency detection model based on inter image saliency propagation and intra image manifold ranking is proposed to fully exploit the saliency bootstrap propagation mechanism of single image, and improve the accuracy of the co saliency detection algorithm. For any pair of images in a group image containing N images, the first inter saliency propagation stage utilizes the similarity between a pair of images to discover common properties of the images and get N-1 initial co saliency maps with the help of a single image saliency map. In order to effectively suppress the background disturbance, the efficient manifold ranking（EMR）algorithm is used to calculate the ranking scores of each initial co saliency maps in the second stage. The ranking scores are then directly assigned to all pixels as their new saliency values. Finally, an integration algorithm is proposed in the Bayesian framework to get the final co saliency map. Based on iCoseg and MSRC image databases, the experimental results show that the proposed algorithm is superior to the five existing co saliency detection algorithms uniformly in F measure and the area under ROC curve (AUC) value. The algorithm is further validated by the real context experiment from the general practicability principle.

查看全文查看/发表评论下载PDF阅读器