Texture-based Image Retrieval Based on FABEMD

In this paper, we present a new method for content based images retrieval (CBIR). We propose characterizing images by using global information extracted from the Fast and Adaptive Bidimensional Empirical Mode Decomposition (FAEMD), which decomposes image into a set of functions named Bidimensional Intrinsic Mode Functions (BIMF) and a residue. On the first two BIMFs, which contains a high frequency part of the image, eventually curves and edges, Curvelet transform (CT) was applied; wheras on the remaining part of the image Gabor wavelets (GW) were applied. Image feature based on Curvelet transform and Gabor wavelet, are then calculated. Our approach was tested on Brodatz database. Experimental results show that the proposed system outperforms previous rotation-invariant systems significantly, and it is found to be superior to Curvelet Transform and Gabor wavelets.