A Framework for Document Pre-Processing in Forensic Handwriting Analysis

analysis Katrin Franke, Mario K oppen Fraunhofer Institute for Produ tion Systems and Design Te hnology IPK, Pas alstr. 8-9, D-10587 Berlin, Germany, Fon: +49 30 39006194, Fax: +49 30 3917517, E-mail: fkatrin.franke, mario.koeppeng ipk.fhg.de We propose an open layered framework, whi h might be adapted to ful ll sophisti ated demands in forensi handwriting analysis. Due to the ontradi ting requirements of pro essing a huge amount of di erent do ument types as well as providing high quality pro essed images of singular do ument lasses, neither a standardized queue of pro essing stages and xed parameter sets nor xed image operations are quali ed for su h a framework on ept. The open layered framework, proposed in this paper, provides adaptation abilities at the parameter level, the operator level and the algorithm level. Moreover, an embedded module that uses geneti programming might generate spe i lters for ba kground removal on the y. In the following the layered framework will be presented, aspe ts of the implementation and results of its appli ation will be given. 1 Introdu tion Forensi handwriting examination, usually performed on request of the justi e or private persons, deals with the analysis and the evaluation of handwriting to dete t frauds. A ording to the la k of obje tive measurements and reprodu ible de isions, traditional methods, like visual inspe tion and expert rating, were tried to repla e by omputerized semi-automati and intera tive systems. Two systems operating in forensi labs for su h a purpose are the Fish-system1 and the S ript-system2. Computer supported investigation of handwriting, using methods of digital image pro essing and pattern re ognition, requires sophisti ated pre-pro essing fun tions to remove ba kground textures of do uments and other imprints, whi h are not subje t of the examination. Unfortunately, at the beginning of our proje t in 1997, there wasn't a suitable solution available, whi h was able to pro ess huge amounts of di erent do ument types as they o ur in daily forensi asework and whi h would provide high quality handwriting images as they are needed for further examinations3. Furthermore, major drawba ks were aused by only partially eliminated noise signals, ases where a orre t separation of handwriting from textured ba kground was impossible and ases where parts of the handwriting were lost, in parti ular those with a low on 73