【摘要】 现有显著性检测方法大多存在检测误差大、主观性强、对背景先验知识约束过少等局限性。为此,提出一种背景检测显著性同步更新优化方法。通过改变背景先验的约束范围,计算显著图与真值图的相似程度,利用置信度量进行同步传播更新,使相邻像素间的关联性得以加强,显著目标边缘更清晰。在标准数据集上的实验结果表明,与现有基于背景的显著性检测方法相比,优化方法具有更高的检测精度。更多还原

【Abstract】 The existing saliency detection methods have many limitations,such as large detection error,strong subjectivity and too little restriction on the background prior knowledge. So this paper proposes a synchronization update optimization method based on the saliency of backgrounddetection. Through changing the constraints of background prior,the similarity degrees of significant graphs and truth graphs are calculated,and the confidence metric is used to synchronize the propagation update. This method strengthens the correlation between adjcent pixels and makes significant target edge clearer. Experimental results on the standard dataset showthat,the optimization algorithm significantly improves the detection accuracy compared with existing background-based saliency detection methods.更多还原