Mesure de la netteté sur une image seule dans des documents anciens

This article presents some way to do quality control after digitization, specially out of focus problem. We will enumerate different point of view to analyse and to estimate this information. To validate descriptors, we suggest to blur any picture and to compute blur estimation in order to evaluate precision. After that, we propose to combinate descriptors by machine learning. MOTS-CLÉS : documents anciens, apprentissage automatique, flou de focus, contrôle qualité, workflow de numérisation.