Bootstrap Empirical Mode Decomposition with Application to CIELAB Color Images

Abstract

This paper proposes an alternative optimization-based EMD based on the notions of: 1. <i>local mean points </i>that impose mode symmetry via a Tikhonov regularized least-square (RLS) problem, and 2. efficient <i>bootstrap sifting</i> that guarantees asymptotic convergence of the mean envelope to the local mean points, regardless of regularization. Mathematical proof of convergence and a straightforward extension to the 2D-multivariate setting and CIELAB color image sare presented. Performance is demonstrated with a univariate signal and two images. Spectral analysis confirms coordinated feature extraction among image components, and separation of spatial spectra among the intrinsic mode functions.