Generalized subspace rules for on-line PCA and their application in signal and image compression (TP-L5)

Author(s) : Toshihisa Tanaka (Tokyo University of Agriculture and Technology, Japan)

Abstract : A general structure of weighted subspace (WS) rules for principal component analysis (PCA) is introduced. Several efficient algorithms for PCA which track principal components with preserving orthogonality without any normalization have been proposed. In this paper, we unify them into generalized forms of the WS rules. By the analysis of these generalized rules, we find the optimal cases of the rules in terms of the preservation of orthogonality of estimated principal components during tracking. The results of theoretical analysis on the stability are shown. To understand the theoretical behavior, them, toy numerical examples are shown. Moreover, a possibility for the application of adaptive data compression is discussed, by showing examples of backward adaptation image coding.

Menu