Removing redundancy data with preserving the structure and visuality in a database
Removing redundancy data with preserving the structure and visuality in a database
复制标题
DOI：
10.1007/s11760-018-1404-8
复制
发表时间：
2018-12
期刊：
影响因子：
--
通讯作者：
Ali Asghar Sharifi Najafabadi;Farah Torkamani Azar
中科院分区：
文献类型：
--
作者：
Ali Asghar Sharifi Najafabadi;Farah Torkamani Azar
人脸图像等数据库的准备和维护中最重要的挑战之一是原始数据中存在大量变量。减少每个观察结果的大小，使其特征不会消失，并且共同签名者唯一地提供对原始数据的依赖性，这是数据库运行的必要要求。主成分分析及其改进方法是降低此类数据集维数，同时提高可解释性和最小化信息损失的常用技术。在本文中，我们的目的是定义一个预处理步骤作为非均匀采样，以保持原始数据的外观，并提高降维方法的性能。我们使用稀疏主成分分析的属性来识别原始数据中不太重要的值的位置，这些值不会干扰数据特征。通过使用稀疏特征向量，提出了两种算法，以消除冗余的原始数据在一维和二维的情况下。在去除原始数据冗余之后，可以使用在其他应用（例如数据库识别和压缩）中新获得的数据。仿真结果表明，采用该预处理步骤可以减少存储量，并提供更高的识别率。
One of the most important challenges in the preparation and maintenance of databases such as face images is the presence of a large number of variables in the raw data. Reducing the size of each observation so that its features do not disappear and the co-signer uniquely provides dependence on the original data is a necessary requirement of functioning with databases. Principal component analysis and its improved methods are the common techniques for reducing the dimensionality of such datasets and simultaneously increasing the interpretability and minimizing information loss. In this paper, our purpose is to define one preprocessing step as nonuniform sampling to preserve raw data appearance and to increase the performance of the dimensional reduction methods. We use the properties of sparse principal component analysis to identify the the location of less important values of the raw data that do not interfere with data features. By using sparse eigenvectors, two algorithms are presented to remove redundancy from the raw data in the one-dimensional and two-dimensional cases. After removing raw data redundancy, newly obtained data in other applications such as database recognition and compression can be used. Simulation results show that using this preprocessing step reduces the memory amount and also provides the higher recognition rate.