Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (2): 179-186.DOI: 10.3778/j.issn.1002-8331.1710-0208

Previous Articles     Next Articles

Salient Object Detection Algorithm via Multiple Prior Fusion

DONG Benzhi, YU Shangshu, JING Weipeng   

  1. College of Information and Computer Engineering, Northeast Forestry University, Harbin 150040, China
  • Online:2019-01-15 Published:2019-01-15

多先验融合的图像显著性目标检测算法

董本志,于尚书,景维鹏   

  1. 东北林业大学 信息与计算机工程学院,哈尔滨 150040

Abstract: In order to detect the salient object more accurately, a new salient object detection algorithm based on multiple prior fusion is proposed. Traditional center prior failed to detect salient object deviated from the center of image, the minimum convex hull is got by using the intersection of multi color space, and it can determine the location of the object and compute center prior by convex hull region. At the same time, a fusion strategy is proposed, which integrates the convex hull region center prior, color contrast prior and background prior into feature matrix. Finally, the saliency map is generated by the low rank matrix recovery model. The simulation experiments on the open dataset MSRA1000 and ESSCD show that MPLRR can obtain clear and significant salient object visual effect map. At the same time, F, AUC, MAE and other evaluation indicators are also significantly improved than many existing methods.

Key words: Multiple Prior fusion Low Rank Matrix Recovery(MPLRR) algorithm, salient object, convex hull region center prior, fusion strategy, low rank model

摘要: 为了更加准确地检测出图像中的显著性目标,提出了多先验融合的显著性目标检测算法。针对传统中心先验对偏离图像中心的显著性目标会出现检测失效的情况,提出在多颜色空间下求显著性目标的最小凸包交集来确定目标的大致位置,以凸包区域中心计算中心先验。同时通过融合策略将凸包区域中心先验、颜色对比先验和背景先验融合并集成到特征矩阵中。最后通过低秩矩阵恢复模型生成结果显著图。在公开数据集MSRA1000和ESSCD上的仿真实验结果表明,MPLRR能够得到清晰高亮的显著性目标视觉效果图,同时F,AUC,MAE等评价指标也比现有的许多方法有明显提升。

关键词: MPLRR算法, 显著性目标, 凸包区域中心先验, 融合策略, 低秩模型