Abstract:

We present a 3-fold distance metric and a transfer function to evaluate the similarity of two finite length sequences. We analyze the sensitivity characteristics of the proposed metrics for Gaussian shape functions. Our method is based on cross-correlation matrix analysis and extrapolation of a minimum cost path using dynamic programming. Unlike the existing sequential (bin-by-bin) and non-sequential (cross-bin) approaches that compute a single scalar as a result of the measurement, we calculate the distance as well as determine how two sequences are correlated with each other in terms of a non-parametric transfer function. We shown that the proposed metrics provide better discrimination than conventional metrics do. Furthermore, we show that we can reduce our metric to any one of sequential metrics with suitable simplification.