A green’s function-based Bi-dimensional empirical mode decomposition

Abstract

Bidimensional Empirical Mode Decomposition(BEMD) interprets an image as a superposition of Bidimensional Intrinsic Mode Functions (BIMFs). They are extracted by a process called sifting, which encompasses two-dimensional surface interpolations connecting a set of local maxima or minima to form corresponding envelope surfaces. Existing surface interpolation schemes are computationally very demanding and often induce artifacts in the extracted modes. This paper suggests a novel method of envelope surface interpolation based on Green’s functions. Including surface tension greatly improves the stability of the new method which we call Green’s function in tension-based BEMD (GiT-BEMD). Simulation results, using toy images with various textures, facial images and functional neuroimages, demonstrate the superior performance of the new method when compared to its canonical BEMD counterpart. GiT-BEMD strongly speeds up computations and achieves a higher quality of the extracted BIMFs. Furthermore, GiT-BEMD can be extended simply to an ensemble-based variant (GiT-BEEMD), if needed. In summary, the study suggests the new variant GiT-BEMD as a highly competitive, fast and stable alternative to existing BEMD techniques for image analysis.