Manifold Regularized Low Rank Embedding for Hyperspectral Image Feature Extraction
Recently,low rank embedding(LRE)method has achieved great success in robust image feature extraction,which aims to embed the data into a low dimensional space with the low rank reconstruction relationship preserved.Since the high dimensional data of hyperspectral image(HSI)often leads to information redundancy,LRE is considered to perform the feature extraction of HSI in this paper.Although LRE can seek the low rank representation(LRR)and optimal subspace simultaneously,the characteristic of LRR results in that the subspace obtained by LRE only considers the global Euclidean structure and ignores the local manifold structure.However,the local manifold structure generally plays a important role for HSI robust features extracting.In order to exploit the local manifold structure of the data,a Laplacian graph characterized manifold regularization has been incorporated into LRE,leading to our proposed Laplacian regularized LRE(LapLRE).Classification by a existing classifier is implemented to verify the robustness of features extracted by LapLRE.Experimental results on two HSI data sets demonstrate that the performance of LRE has been enhanced by using the manifold regularization.
low rank embedding hyperspectral image feature extraction manifold regularization classification
Heng-Chao Li Yang-Jun Deng Wen Yang
School of Information Science and Technology,Southwest Jiaotong University,Chengdu 610031,China School of Electronic Information,Wuhan University,Wuhan 430072,China
国内会议
武汉
英文
1-9
2017-09-17（万方平台首次上网日期，不代表论文的发表时间）