Spectral characterization of orientation data along curvilinear structures

This paper lies within the scope of texture analysis. We focus on anisotropic textures and propose an approach to measure the ripple of texture patterns. It assumes that the patterns are represented by some characteristic curves called generatrices (i.e. edges, level curves). Two steps are involved. The first one deals with the estimation of the orientation along the generatrices. The second step consists in computing the power spectrum of the generatrix orientation. The global texture ripple is captured by the ripple spectrum which is computed by cumulating the orientation spectra of all individual generatrix. The approach is exercised on synthetic textures and on composite material images.