摘要
针对当前显著性检测算法普遍存在的背景噪声较多,目标区域检测不够完整等问题,提出了一种空间域的显著性算法.首先将输入图像进行超像素分割,以边缘信息作为背景先验区域集,通过计算超像素与背景先验区域集内超像素在颜色、亮度方面的差异,得到背景差异显著图;然后确定前景先验区域集,计算各超像素与前景先验区域集内超像素的差异性,得到前景差异显著图.最后融合两部分显著图;最后在此基础上构建视觉中心,围绕视觉中心确定各超像素空间权重信息,得到最终显著图.采用MSRA-1000数据库进行对照实验,结果表明本文算法的准确性更高,整体效果更好.
Aiming at the problems that the current saliency detection algorithm generally has more background noise and the target area detection is not complete enough,a spatial domain saliency algorithm is proposed.Firstly,the input image is superpixel segmented,and the edge information is used as the background prior area set.By calculating the difference in color and brightness between the superpixel and the superpixel in the background prior area set,a significant background difference map is obtained.Then it is determined the foreground a priori area set,calculated the difference between each superpixel and the superpixel in the foreground a priori area set,and obtained a foreground difference saliency map.Finally,it is fused the two parts of the saliency map,builded a visual center on this basis,and determined the spatial weight information of each superpixel around the visual center to obtain the final saliency map.The MSRA-1000 database was used for a comparative experiment.The results show that the algorithm in this paper is more accurate and the overall effect is better..
作者
邵凯旋
余映
钱俊
吴青龙
杨鉴
SHAO Kai-xuan;YU Ying;QIAN Jun;WU Qing-long;YANG Jian(School of Information Science and Engineering,Yunnan University,Kunming 650500,China)
出处
《云南大学学报（自然科学版）》
CAS
CSCD
北大核心
2020年第3期429-436,共8页
Journal of Yunnan University(Natural Sciences Edition)
基金
国家自然科学基金(61263048)
云南省应用基础研究计划(2018FB102).
关键词
显著图
超像素分割
前景预测
颜色权重
saliency map
superpixel segmentation
edge information
spatial weight
