摘要
针对环境复杂的图像组中前景和背景信息混乱、共显性目标无法准确显示的问题,提出一种特征融合和多重约束的图像协同显著性检测算法.首先通过融合对象性概率优化的深度信息和颜色信息计算图像间显著性引导传播的显著值;然后使用深度概率指导的多重约束的背景先验方法计算图像内显著值以进一步优化;将两阶段得到的显著值进行区域性建议融合,采用最小二乘法学习得到最终的协同显著性结果.在公共数据集上的实验结果表明,该算法有效地利用图像间相似性信息且抑制了背景信息,使得显著目标更加接近真值标定结果;克服了复杂环境因素的影响,在各个实验指标上的评估结果都有明显提高.
In order to solve the problem of the confusion of foreground and background information and the wrong indication of salient objects for image groups with complex environments,we propose a co-saliency detection model based on feature fusion and multiple constraints.Firstly,we calculate the inter-aliency values by fusing depth information optimized by objectness probability and color information.Then,we calculate the intrasaliency values by background priors with multiple constraints guided by depth probability.The saliency maps are also optimized.Finally,we fuse the saliency maps with region-wise proposal.The least squares method is used to obtain the final results.The experimental results on public datasets indicate that the consistent relationship among multiple images is used more effectively and background information is better suppressed.The salient objects are closer to the truth values.Overcoming the impact of complex environmental factors,the evaluation results of this algorithm have been significantly improved.
作者
金志刚
李静昆
Jin Zhigang;Li Jingkun(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2019年第9期1477-1484,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61503084)
关键词
协同显著性检测
特征融合
多重约束
深度信息
co-saliency detection
feature fusion
multiple constraint
depth information
