基于鲁棒前景选择的显著性检测

发布时间：2018-02-25 07:16

本文关键词： 显著性检测凸包锚点图流形排序　出处：《电子与信息学报》2017年11期 　论文类型：期刊论文

【摘要】：显著性检测是指自动提取未知场景中符合人类视觉习惯的兴趣目标的方法。为了进一步提高检测的准确性,该文提出了利用鲁棒前景种子的流形排序进行显著性检测的算法。首先利用角点检测和边缘连接算法得到两个不同的凸包,用它们的交集初步确立目标区域的大致位置;然后利用凸包外边缘作为标准对凸包内的超像素进行相似度检测,将与大部分外边缘相似的超像素去除,得到更准确的目标样本作为前景种子;利用锚点图构建新的图结构表示数据节点之间的关系;接着通过基于前景和背景种子的流形排序算法对图像所有区域进行排序,并得到两种不同的显著性检测图;最后借助代价函数对显著性图进行优化,得到最终的显著性检测结果。经实验表明,与几种经典算法对比,该文方法可以进一步提高显著性算法的精确度和召回率。
[Abstract]:Saliency detection refers to the automatic extraction method in line with the human visual habits in the unknown target in the scene. In order to further improve the detection accuracy of proposed saliency detection algorithm using robust foreground seed manifold ranking. Firstly by corner detection and edge linking algorithm to obtain two different hull, preliminary establishment the approximate location of the target area by their intersection; then the outer edge of the convex hull as a standard within the convex hull super pixel similarity detection, will be similar to most of the outer edges of the super pixel removal, get more accurate target samples as seed; using anchor graph construction graph structure of new relational data between nodes; then through the manifold ranking algorithm based on foreground and background seeds for all areas of the image are sorted, and get two significantly different after the detection map; The saliency map is optimized by cost function, and the final saliency detection result is obtained. Experiments show that compared with several classical algorithms, this method can further improve the accuracy and recall rate of saliency algorithm.

【作者单位】：西北工业大学电子信息学院;空军工程大学航空航天工程学院;
【基金】：国家自然科学基金(61379104)~~
【分类号】：TP391.41

【相似文献】