文献类型：期刊文献

中文题名：面向社交媒体图像的显著性数据集

英文题名：A saliency dataset for social images

作者：梁晔[1,2];马楠[1];郎丛妍[2];于剑[2]

第一作者：梁晔

机构：[1]北京联合大学机器人学院;[2]北京交通大学计算机与信息技术学院

第一机构：北京联合大学机器人学院

年份：2018

卷号：42

期号：5

起止页码：135-140

中文期刊名：北京交通大学学报

外文期刊名：Journal of Beijing Jiaotong University

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2017_2018】；

基金：国家自然科学基金(61871038;61871039);北京市自然科学基金(4182022);北京市属高校高水平教师队伍建设支持计划项目(IDHT20170511)~~

语种：中文

中文关键词：数据集;显著性;社交媒体图像;标签

外文关键词：dataset;saliency;social images;tags

摘要：随着显著性研究的发展,已涌现多个显著性数据集,然而目前面向社交媒体图像的显著性数据集数量非常少.为此构建此类显著性数据集,详细论述了数据集的图像来源、图像的筛选原则、图像的标注及数据集的统计分析.为了验证新建数据集的性能,与目前流行的7个显著性数据集进行性能评测,新建数据集具有显著区域尺寸丰富、与图像边界连接度高、显著区域与图像的颜色差异小的优点.实验结果表明:新建数据集中显著区域与图像边界连接的比例为17%,仅低于ECSSD数据集;其中显著区域和整幅图像的颜色差均值最小,且包含10个尺寸等级的显著区域,尺寸分布最广泛.此外,新建数据集具有标签信息,也为新的显著区域提取方法提供了实验对象.
With the development of saliency research, there have been a number of saliency data sets, but saliency datasets for social images are few. Aiming at the current situation, a saliency dataset for social images is constructed. The source of images, principles of image selection, an notation of images and statistical analysis of dataset are discussed in detail. In order to verify the performance of the new dataset, the new dataset and the current seven popular datasets are evalu ated. The new dataset has the advantages of rich salient region sizes, high connectivity with image boundaries, small color difference of salient regions and images. In the new dataset, the ratio of salient regions connected with image boundaries is 17%, which is only lower than that in the ECSSD dataset; the mean color difference between salient regions and whole images is the smallest in the eight experimental datasets; salient regions have 10 size grades which are the widest size distribution. In particular, the new dataset has tag information and provides experimental dataset for new saliency detection methods.