Optical imaging is a non invasive way to characterise turbid media, which is of real interest for investigating biological tissues for diagnosis purposes. A method called Integral Reflectance (IR) has already been developed . The media being illuminated by a laser beam (670 nm, <1 mW), the backscattered light is captured by a 2D CCD camera. The reduced scattering coefficient μ's and the absorption coefficient μa are determined from the image. Having μ's and μa, the objective is to improve the characterization by estimating the anisotropy factor g, using polarized light. Different patterns depending on g are produced in these images, presenting some lobes, centred in the entry point of the laser beam, whose number and shape vary with g. To assess a simple description of these patterns, a circular outline of the image, at a given radius, is studied by Fourier series decomposition, namely Fourier descriptors, whose indices, modulus and phase provide the number, the size and the orientation of the lobes, respectively. Backscattered images of turbid media with g in the range [0.006 ; 0.93] (μ's = 10, 20, 40 cm-1 ; μa = 0.01, 1, 5 cm-1), were simulated using a Monte Carlo code for polarized light. Tables of Fourier descriptors were obtained as function of g, μ's and μa. Five reference solutions made of polystyrene spheres in liquid, with g varying from 0.71 to 0.919 (tissue phantoms) were tested. The Fourier descriptors were compared to simulations, and g could be retrieved with a maximum error of 10%.