Multiple Multidimensional Sequence Alignment Using Generalized Dynamic Time Warping
Parinya Sanguansat · 2012
Nowadays, the alignment of multidimensional sequences is required in many applications especially for multimedia data. The ordinary DTW is not well satisfied this condition because it can only align a pair of 1D sequences at once. Applying traditional DTW to these tasks not only makes each pairwise alignments are independent but also causes each dimensions are compared separately. In this paper, the generalized algorithm of Dynamic Time Warping (DTW) is proposed for multiple multidimensional time series alignments, called Multiple Multidimensional Dynamic Time Warping (MM-DTW). This algorithm utilizes all dimensions to obtain the optimal path and aligns multiple signals simultaneously. After alignment, the regular similarity measurement can be used to these aligned signals. The performances of MM-DTW are investigated by the nearest neighbor classifier compared to the ordinary DTW and its multidimensional modified algorithm. Experiments on real-world application, Query by humming, demonstrate the improved performance of the proposed method over the other algorithms.