In this paper we present the definition and framework of Directional Empirical Mode Decomposition (DEMD) and use DEMD to do texture segmentation. As a new technique of time-frequency analysis, EMD decomposes signals b...In this paper we present the definition and framework of Directional Empirical Mode Decomposition (DEMD) and use DEMD to do texture segmentation. As a new technique of time-frequency analysis, EMD decomposes signals by sifting and then analyzes the instantaneous frequency of the obtained components called Intrinsic Mode Functions (IMFs). Compared with Bidimensional EMD (BEMD) which only extracts textures by radial basis function interpolation, the virtues of DEMD include: the directional quality is considered in this framework; four features can be extracted for each point from the decomposition. The technique of selecting directions for DEMD based on texture’s Wold theory is also presented. Experimental results indicate the effectiveness of the method for texture segmentation. In addition, we show the explanation for the DEMD’s ability for texture classification from visual views.展开更多
文摘In this paper we present the definition and framework of Directional Empirical Mode Decomposition (DEMD) and use DEMD to do texture segmentation. As a new technique of time-frequency analysis, EMD decomposes signals by sifting and then analyzes the instantaneous frequency of the obtained components called Intrinsic Mode Functions (IMFs). Compared with Bidimensional EMD (BEMD) which only extracts textures by radial basis function interpolation, the virtues of DEMD include: the directional quality is considered in this framework; four features can be extracted for each point from the decomposition. The technique of selecting directions for DEMD based on texture’s Wold theory is also presented. Experimental results indicate the effectiveness of the method for texture segmentation. In addition, we show the explanation for the DEMD’s ability for texture classification from visual views.
