Spatial-variant Image Filtering Based on Bidimensional Empirical Mode Decomposition

This paper presents a fully automatic spatial-variant approach for image filtering and representation based on bidimensional empirical mode decomposition (BEMD). Unlike traditional filtering strategies which demonstrate poor performance for multicomponent, non-stationary images, the proposed method adaptively tracks the local characteristics of image intensities. In this paper, we first describe our own BEMD algorithm and use it to decompose gray level images into a finite number of spatial frequency components, called intrinsic mode functions (IMF). Then based on the statistical properties of the IMFs, features can be extracted. The idea is to group certain adjacent modes together to realize image filtering. Experiments on natural multipartite images have indicated the effectiveness of our approach