-Dans cet article, nous proposons une approche originale de la caractérisation des écritures
The aim of this scientific work is to propose an original approach of writers characterization based on the handwriting multi scale decomposition into two main features: curvature and orientation. For every shape point, those two dimensions are extracted by a Curvelets analysis before getting joined together in a compact signature. Curvelets are a specialised version of anisotropic wavelets which are well adapted to the representation of discontinuities along shapes. This is a new geometric multi scale transform for which atoms are indexed by their position, scale and orientation. They integrate the concept of directionnality and allow a sparse representation of images containing objects with many borders, as it is the case of handwritings. For each image, this characterization is synthesized in a compact and single signature used in our information retrieval system dedicated to medieval and humanistic writings. This tool provides very promising results to the use of experts in histories, literatures and palaeographers.