The paper presents a document analysis system to retrieve metadata from digitized ancient manuscripts. This platform has been developed to assist re-searchers, historians and libraries to process a wide variety of manuscripts written in different languages. In order to retrieve different metadata from vari-ous digitized documents, we propose a user-training system, which use robust approaches based on a sequential bottom-up process. We develop a low-level segmentation and a basic recognition stage which do not use prior knowledge on documents contents. Our objective was to study the feasibility to process a large variety of manuscripts with the same platform, which can be used by non-specialists in image analysis.