Image analysis based on an improved bidimensional empirical mode decomposition method

Abstract

The Empirical Mode Decomposition (EMD) is a new adaptive signal decomposition method, which is good at handling many real nonlinear and nonstationary one dimensional signals. It decomposes signals into a a series of Intrinsic Mode Functions (IMFs) that was shown having better behaved instantaneous frequencies via Hubert transform (The EMD and Hubert spectrum analysis together were called Hilbert-Huang Transform (HHT) which was proposed by N.E. Huang et al, in [5].). For the advanced applications in image analysis, the EMD was extended to the bidimensional EMD (BEMD). However, most of the existed BEMD algorithms are slow and have unsatisfied results. In this paper, we firstly proposed a new BEMD algorithm which is comparatively faster and better-performed. Then we use the Riesz transform to get the monogenic signals. The local features (amplitude, phase orientation, phase angle, etc) are evaluated. The simulation results are given in the experiments.