Dependence Maximizing Temporal Alignment via Squared-Loss Mutual Information

Temporal alignment of sequences is an important problem with many practical applications such as speech recognition [1, 2], activity recognition [3, 4], temporal segmentation [5], curve matching [6], chromatographic and micro-array data analysis [7], synthesis of human motion [8], and temporal alignment of human motion [9, 10]. Dynamic time warping (DTW) is a classical temporal alignment method that aligns two sequences by minimizing the pairwise squared Euclidean distance [1, 2]. An advantage of DTW is that the minimization can be efficiently carried out by dynamic programming (DP) [11]. However, due to the Euclidean formulation, DTW may not be able to find a good alignment when the characteristics of the two sequences are substantially different (e.g., sequences have different amplitudes). Moreover, DTW cannot handle sequences with different dimensions (e.g., image to audio alignment), which limits the range of applications significantly.