基于散度-形状引导和优化函数的显著性目标检测

为了准确地进行显著性目标检测,本文提出了一种基于散度-形状引导和优化函数的显著性检测有效框架。首先,通过考虑颜色、空间位置和边缘信息,提出了一种有辨别力的相似性度量。接着,利用散度先验剔除图像边界中的前景噪音获得背景集,并结合相似性度量计算得到基于背景显著图。为了提高检测质量,形状完整性被提出并通过统计在分层空间中区域被激活的次数期望生成相应的形状完整显著图。最后,利用一个优化函数对两个显著图融合后的结果进行优化从而获得最终的结果。在公开数据集 ASD、DUT-OMRON和ECSSD上进行实验验证, 结果证明本文方法能够准确有效地检测出位于图像任意位置的显著性物体。

Saliency detection based on scatter-shape guidance and optimization function

In order to detect saliency object accurately, this paper proposes an efficient framework for saliency detection based on scatter-shape guidance and optimization function. First, it proposed a discriminative similar metric by taking color, spatial and edge information into consideration. Based on similar metric together with background set obtained by removing the foreground noise in the image boundaries with scatter-guided, it constructed a background based saliency map. In order to improve the quality of detection, it introduced the shape completeness cue to generate the corresponding shape completeness saliency map by measuring the completeness of a region by the expectation of times for which the region is bounded by completely shape over the hierarchical space. Finally, it achieved the final saliency map by integrating the above both maps jointly into an optimization function. Quantitative experiments on four available datasets ASD, DUT-OMRON and ECSSD demonstrate that the proposed method outperforms other state-of-the-art approaches and detects the salient object which locates at random positions.