Sequences feature vectors extracting method for similarity measurement based on wavelet and matrix transforming

Abstract

A feature vectors extracting method for similarity measurement between a referenced sequence and an analyzed sequence is proposed. The referenced sequence and analyzed sequence are compressed into two wavelet matrices by Discrete Orthogonal Wavelet Transform (DOWT), respectively. A singular value vector and the multi-subspaces of the referenced matrix are derived from wavelet matrices by singular value decomposition (SVD). Consequently, a uniform subspace of which all sequences are mutual orthogonal can be constructed by serializing multi-subspaces, and the analyzed feature vectors can also be obtained by inner product transformation between analyzed sequence and all sequences derived from the multi-subspaces. The similarity is measured between the analyzed feature vector and the singular value vector of the referenced sequence. The simulation results show that the proposed method is improved in the dimension, accuracy and anti-noise ability with little sensitivity sacrifice.