融合流形排序和能量方程的显著性检测算法  

Saliency Detection Algorithm Combining Manifold Ranking and Energy Equation

作　　者：任守纲[1,2] 陆海飞 徐焕良[1,2] 

机构地区：[1]南京农业大学信息科技学院,南京210095 [2]江苏省肉类生产与加工质量安全控制协同创新中心,南京210095

出　　处：《小型微型计算机系统》2016年第9期2125-2129,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61403205)资助;国家科技支撑计划项目(2015BA1105000)资助;江苏省自然科学基金项目(BK2012363)资助;江苏省博士后科研计划项目(1302038B)资助;江苏省农业三新工程项目(SXGC2014309)资助

摘　　要：显著性检测是当前机器视觉研究的重要问题,针对context-aware(CA)算法在检测过程中造成边缘丢失且易造成冗余检测的问题,提出了一种融合流形排序和能量方程的显著性检测算法(MREESD).该算法使用超像素方法将图像分块,提出一种新的超像素间权重计算方法和显著种子选取方法,通过一种新的显著度计算方法-流形排序计算显著图,最后利用能量方程对得到的显著图进行调整,对得到的显著图进行阈值分割,得到二值图像,再将二值图像与原图像进行掩码运算,得到分割结果.在MSRA1000图像显著性检测数据库上进行测试,准确率-召回率曲线显示在相同召回率下准确率高于其他算法并且具有较高的F-measure值.该算法有效地解决了边缘丢失及冗余分割问题,而且分割效果更加精确.Saliency detection plays an important role in computer vision. In order to solve the problems of edge loss and over segmenta- tion in context-aware C CA ） algorithm, a new saliency detection algorithm combining manifold ranking and energy equation, called MREESD is proposed. The algorithm uses the superpixel segmentation method to divide the image into blocks, puts a new method for calculating the weights of superpixels and selecting saliency seeds. After combining a new way-manifold ranking computing the sali- ency map, using the energy equation to adjust the saliency map, also using the threshold segmentation for the obtained saliency map, fi- nally the foreground and background of the original image are separated by adding the binary map to the original image. The proposed algorithm has been tested on MSRA1000 image saliency detection database. The precision-recall curve shows that a higher precision and F-measure can be obtained than other methods. The experimental result shows that the proposed algorithm effectively solves the edge loss and over segmentation and it＇s more accurate in segmentation.

关 键 词：显著性检测 边缘权重 流形排序 显著种子 能量方程 图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]