A Haar wavelet-based multi-resolution representation method of time series data

Similarity search of time series can be efficiently handled through a multi-resolution representation scheme which offers the possibility to use pre-computed distances that are calculated and stored at indexing time and then utilized at query time together with filters in the form of exclusion conditions which speed up the search. In this paper we introduce a new multi-resolution representation and search framework of time series. Compared with our previous multi-resolution methods which use first degree polynomials to reduce the dimensionality of the time series at different resolution levels, the novelty of this work is that it applies Haar wavelets to represent the time series. This representation is particularly adapted to our multi-resolution approach as discrete wavelet transforms have the ability of reflecting the local and global information content at every resolution level thus enhancing the performance of the similarity search algorithm, which is what we have shown in this paper through extensive experiments on different datasets.