Title :
Metrics of grassmannian representation in reproducing kernel hilbert space for variational pattern analysis
Author :
Washizawa, Yoshikazu
Author_Institution :
Univ. of Electro-Commun., Chofu, Japan
Abstract :
Variation of patterns in signal can be represented by the covariance structure of vectors or its eigensubspace. When information of the pattern variation is available, representation by the covariance matrix or the eigensubspace is useful for feature extraction and classification compared with standard vector or matrix representations.
Keywords :
Hilbert spaces; covariance matrices; eigenvalues and eigenfunctions; feature extraction; signal representation; covariance matrix; covariance structure; eigensubspace; feature extraction; grassmannian representation; kernel Hilbert space; matrix representations; variational pattern analysis; Correlation; Kernel; Manifolds; Measurement; Principal component analysis; Standards; Training; Grassmann manifold; Mahalanobis distance; Subspace distance; kernel principal component analysis; kernel trick;
Conference_Titel :
Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on
Conference_Location :
South Brisbane, QLD
DOI :
10.1109/ICASSP.2015.7178360
