A Nonlinear Differential Equation for Generating Warping Function

Abstract

Given set of functions $y_i(t)$ and $x(t)$ such that $y_i(t) = a_i x\left[h_i(t)\right]$ with $a_i$ being an unknown amplitude with low changes in time (or $\frac{\Delta a_i}{a^2_i} << 1$) and $h_i(t)$ an unknown warping function, the paper shows that $h_i(t)$ can be described using a non-linear differential equation. The differential equation then can be utilized to estimate the warping function $h_i(t)$ using a nonlinear least-squares optimization. This differential equation can also be useful for reducing and analyzing phase variability in data sequences. Results, obtained on synthetic curves, showed that the proposed method is effective in aligning the curves. The obtained aligned curves exhibit variation only in amplitude, and phase variation can be removed efficiently.