摘要
图像是一种高维数据,在图像检索中容易产生维数灾难问题。传统的降维方法很难有效地揭示高维数据的内在本质结构,而流形学习是一种非线性降维方法,其目的是获取高维观测数据的低维嵌入表示并挖掘出隐藏在高维图像数据中的本征信息与内在规律。本文结合SIFT特征提取算法与ISOMAP流形学习算法在人脸图像数据集上进行检索实验。分析探讨近邻参数以及内在本征维数的大小对人脸图像识别效果的问题。
Image data is high-dimensional data which make it easily prone to the dimension disaster. The traditional dimensionali-ty reduction methods can not recover the inherent structure. Manifold learning is a nonlinear dimensionality reduction technique, it aims to find low-dimensional compact representations of high-dimensional observation data and explore the inherent law and in-trinsic dimension of data. In this paper, the feature extraction method-SIFT and the adaptive ISOMAP method are combined and conducted on the real face image dataset. This paper analyzes and discusses the problem of the effects of the neighborhood param-eter and the intrinsic dimension size on the face image recognition.
出处
《计算机与现代化》
2015年第9期25-29,共5页
Computer and Modernization
基金
江苏省高校自然科学研究项目(14KJD520003)
关键词
图像检索
流形学习
降维
本征维数
image retrieval
manifold learning
dimensionality reduction
intrinsic dimension
