A texture descriptor for textile image retrieval

This paper proposes a new descriptor for textures, obtained by an Independent Components Analysis (ICA) applied to image wavelet subbands. In certain conditions, ICA refines the wavelet multiband analysis. The descriptor, consisting in the independent components (ICs) distribution parameters, serves as basis for the textile image signature in a recognition application. By using the Kullback-Liebler divergence to compare the signatures, the textile images from a 30 jacquards databasis, were recognized with average rates ranging from 89 to 94%, depending on the analysis number of levels.