Texture Retrieval by Scale and Rotation Invariant Directional Empirical Mode Decomposition

In this paper, a method for texture retrieval based on scale and rotation invariant directional empirical mode decomposition (SRIDEMD) is presented. Different from other filtering based techniques such as wavelet and Gabor decomposition, EMD uses the nonlinear filtering process called ´sifting´ which attains its scalead aptivity and obtained intrinsic mode functions (IMFs) which has approximate orthogonality. We extend DEMD which is a fast technique of extending 1D EMD to 2D case by introducing scale and rotation invariance. Features including frequency and envelopes of IMFs are extracted after 2D Hilbert transform. Decomposition in several directions is made for rotation invariance and main direction is used. Scale-invariant features are attained by further processing the results and using fractal dimensions of the residues and IMFs. We validate the effectiveness of this method by experiments for textures from public texture database.