背景感知的显著性检测算法

Background aware saliency detection

显著性检测算法能够提取场景中的重要目标,因而在图像和视频处理领域有着重要应用。但目前的显著性检测算法几乎都以目标与背景之间存在显著差异为前提,利用目标与背景的差异以及目标本身的空间相关性进行算法的设计。本文则从图像的背景信息出发,首先利用超像素算法对图像进行预分割,去除图像中不必要的细节信息;然后利用背景信息获得初步的显著性检测结果;最后将多尺度显著性检测结果进行融合,得到最终的显著性图。实验结果表明,本文算法的性能远优于现有检测算法。

Saliency detection can abstract important targets,so it has important application in the field of image and video processing.But almost all the current saliency detection algorithms are based on significant con-trast between the target and the background,and they compute saliency with differences between targets and backgrounds,and space correlation within targets.A new method is provided using the background infonma-tion.First,an image is decomposed into compact elements that abandon unnecessary details.Then the initial saliency map is computed with background information.Finally the multi-scale information is fused to promote the detection performance.A detailed experimental evaluation shows that the proposed method outperforms all state-of-the-art approaches.