基于流形排名与迟滞阈值法的视觉显著性检测

视觉显著性检测是机器视觉领域的关键技术之一.提出一种基于流形排名与迟滞阈值的检测方法,首先将图像划分成超像素集合,以之作为结点形成闭环图;再按照基于图的流形排名方法计算各个结点的显著值,形成图像的显著图;然后利用显著图直方图统计出高、低两个阈值,将显著图划分为三个部分,使用伽马校正技术分别进行处理,最终整合校正结果得到输出显著图.实验结果表明,相对于现有算法,本文算法得到的显著图能够更好地区分背景区域和显著目标,同时也更具稳健性.

Visual saliency detection via manifold ranking and hysteresis threshold

Visual saliency detection is one of the key technologies for machine vision.We propose a new approach based on manifold ranking and hysteresis threshold to detect salient objects in images.Firstly,the image is represented as a close-loop graph with nodes consisting of superpixels,and these nodes are ranked by manifold ranking method in order to extract a saliency map.Secondly we compute two different thresholds from the histogram of saliency map,and apply the gamma correction to this map twice with high and low values respectively.Finally the outputs from gamma corrections are processed by two thresholds so as to segment background regions and thereby salient objects.Experimental results demonstrate that the proposed approach performs well when against the existing algorithms in terms of accuracy and robustness.