Texture segmentation using window empirical mode decomposition

In this paper window empirical mode decomposition (WEMD) is proposed and is used to do texture segmentation. Empirical mode decomposition (EMD) can decompose the nonstationary and nonlinear signals by sifting into a few intrinsic mode functions (IMFs) which represent a simple oscillatory mode of local data. However, the traditional bidimensional EMD (BEMD) has two drawbacks of the gray spots in IMF image and the slow computation speed. WEMD can solve such problems. Based on the characteristic of WEMD and local time/space-frequency analysis of structure multivector, the renovate technique of texture segmentation is also presented. Characterized by the local amplitude and the local frequency of every IMF component, the texture image can be segmented by k-means clustering algorithm. The subsequent experimental results indicate this method's effectiveness.