Self-organizing maps and ancient documents

This paper presents how Self-Organizing Maps and especially Kohonen maps can be applied to digital images of ancient collections in the perspective of valorization and diffusion. As an illustration, a scheme of transparency reduction of the digitized Gutenberg Bible is presented. In this two steps method, the Kohonen map is trained to generate a set of test vectors that will train in a supervised manner a classical feed-forward network. The testing step consists then in classifying each pixel into one class out of four by feeding directly the feed forward network. The pixels belonging to the transparency class are then removed.