A Framework for Cooperative and Interactive Analysis of Technical Documents Maurizio Rigamonti, Oliver Hitz and Rolf Ingold Department of Informatics of University of Fribourg Chemin du Musée 3, CH-1700 Fribourg, Switzerland Phone: +41-26-300 84 79 Fax: +41-26-300 97 31 E-mails: {maurizio.rigamonti, rolf.ingold}@unifr.ch Abstract A powerful user interface is unavoidable for cooperative and interactive document recognizing system. This paper presents xmillum, a framework that allows to visualize and to edit all types of document recognizing data in any XML language. The data to be visualized and edited is transformed into an internal language using an XSL stylesheet. Small modules encapsulate how to render and handle the data. The framework can be freely extended with new modules to improve visualization and editing capabilities. The main contribution of the article is to present xmillum and the advantages of using the framework in document recognition data. Keywords:Document Analysis System,Interaction,System Architecture,XML. 1 Introduction xmillum (XML Illuminator) is a framework for cooperative and interactive analysis of documents developed by the DIVA research group at the University of Fribourg. The philosophy of xmillum is to visualize an object in a graphical context and to offer a set of interactions on it [1, 2]. It is based on the CIDRE 1 philosophy, which promotes the idea that a general document recognition system does not work in a fully automatic way, but cooperates with the user. The feedback of interactions improves the ability of the system to learn incrementally and to adapt itself to different document properties. This approach requires a powerful tool for visualizing and editing. We propose xmillum as a solution for that problem: the framework allows users to correct and validate results of document analysis in real time. It does not require a specific language to represent data and it can be easily extended to analyze many types of documents. The rest of this paper is organized as follow: the second chapter discusses the goals of xmillum; the third chapter presents the architecture of xmillum and how it takes advantage of XML[7] and XSLT[8]; the fourth chapter shows examples of applications of xmillum; the fifth chapter presents the advantages of using the framework; finally, the sixth chapter concludes this paper. 2 Goals of xmillum The visualization of results is very important for everyone working in the document recognition domain. It allows researcher and end-user to evaluate the performance, to analyze and to calibrate their recognition systems. In order to analyze the performance of traditional document recognition systems such as OCR, textual information may be sufficient, but as soon as the recognized data gets more complex, more sophisticated visualization techniques are necessarily. This is true especially in the graphics recognition domain. 1 CIDRE stands for Cooperative and Interactive Document Reverse Engineering and is supported by the Swiss National Fund for Scientific Research, code 2000-059356.99-1, as well as France Télécom R&D