Retrieved Image Refinement by Bootstrap Outlier Test
Retrieved Image Refinement by Bootstrap Outlier Test
复制标题
通过 Bootstrap 异常值测试检索图像细化
DOI：
10.1007/978-3-030-29888-3_41
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Noboru Murata
中科院分区：
文献类型：
--
作者：
Hayato Watanabe;Hideitsu Hino;Shotaro Akaho;Noboru Murata
离群值检测用于识别与给定数据集中的大多数其他数据显著不同的数据点或少量数据子集。使用客观和定量的方法检测异常值具有挑战性。使用统计假设检验框架的方法通过假设特定的参数分布作为数据生成模型而被广泛使用，但是在实际问题中不能保证数据的分布可以通过参数分布来充分近似。本文提出了一种简单的方法，通过假设检验客观地检测离群值，而无需假设离群值分数的特定分布。通过使用任意的离群值得分函数，假设检验用于确定每个给定的样本是否是离群值。假设检验需要检验统计量的分布，并使用bootstrap方法基于给定数据进行估计。通过将其应用于基于文本的图像检索中的离群点检测，验证了所提出的离群点检测方法的有效性，通过去除不相关的图像，提高了图像检索的质量。
Outlier detection is used to identify data points or a small number of subsets of data that are significantly different from most other data in a given dataset. It is challenging to detect outliers using an objective and quantitative approach. Methods that use the framework of statistical hypothesis testing are widely used by assuming a specific parametric distribution as a data generation model, but there is no guarantee that the distribution of data can be adequately approximated by a parametric distribution in practical problems. In this paper, a simple method is proposed to objectively detect outliers by hypothesis testing without assuming a specific distribution of outlier scores. By using an arbitrary outlier score function, hypothesis testing is used to determine whether each given sample is an outlier. The distribution of the test statistics is needed for the hypothesis test, and is estimated based on the given data using the bootstrap method. The effectiveness of the proposed outlier test was verified by applying it to outlier detection for text-based image retrieval, where it improved the quality of image searches by removing irrelevant images.