摘要
传统的一维、二维图像特征提取忽略了图像的结构信息,由此带来识别精度的损失;三维和多维图像的特征提取虽然考虑了数据结构之间的彼此联系,但却带来了维数灾难,增加了计算复杂度。本文利用流形学习,在原始的数据空间中嵌入稳定的流形,从而使多维数据中的特征数据映射到流形上,发现隐含在高维数据集中人们无法感知的低维结构,在不丢失数据信息的前提下,降低原始数据的维数,从而降低计算复杂度。
One-dimensional and two-dimensional image feature extraction ignores the image structure information,which will result in the loss of recognition accuracy. Three-dimensional and multi-dimensional image feature extraction considers the links between each data structure,however it brings the curse of dimensionality,which will increase the computational complexity. Accordingly,in this paper,manifold learning is taken to embed a stable manifold in the original data space,thus the multi-dimensional data feature data can be mapped to the manifold,the invisible low-dimensional structure implicated in the high-dimensional data sets can be perceived,the original data dimension can be reduced without losing data,and thereby the computational complexity can be reduced.
出处
《洛阳理工学院学报（自然科学版）》
2015年第4期59-62,89,共5页
Journal of Luoyang Institute of Science and Technology：Natural Science Edition
基金
国家自然科学基金:基于二维随机映射和一范数优化的有监督图像分类研究(6115200)
关键词
流形学习
特征提取
本征维数
数据挖掘
manifold learning
feature extraction
intrinsic dimension
data mining
