Allographic Biometrics and Behavior Synthesis

This essay deals with the role of allographs in the general framework of the study of writing. It explains why allographs are important (writer individuality), how they can be measured (frequency and contexts (graphic, semantic, non-graphic)) and what applications to expect from them (biometrics, script restoration, artistic creation, cultural history). Résumé Cet essai traite du rôle des allographes dans le cadre général de l’étude de l’écriture. Il explique pourquoi les allographes sont importants (l’individualité de l’écrivain), comment les mesurer (fréquences et contextes graphiques, sémantiques et non-graphiques) et à quelles applications s’attendre (biométrie, restauration de l’écriture, création artistique, histoire culturelle). One of the common elements used to identify a typeface or a handwriting is the shape of the script. Sometimes, however, individual characters may have alternate forms— in those cases the rules and habits governing their usage characterize the behavior of the typesetter or writer in a more subtle way than shape does. Allographs provide inconspicuous signatures to a text. This paper intends to raise the issue of the study of allographs— allometry— , for the pleasure of its intricacies as well as for its many practical applications. I will proceed by discussing four diagrams, representing some of the central issues of the subject. Although my discussion is based on Arabic handwritten examples— a script where allographs are an inherent part of the writing system— , the same principles can be applied to non-Arabic handor machine-produced scripts. Measuring frequencies The two text blocks in Fig. 1 represent a double-page of a manuscript copied in Sicily in the 14th century [1]. On the right side we see the original script (Fig. 1a), whereas the left side (Fig. 1b) is a replacement folio, added at a later date, after some folios went missing. Some attempts were made to keep the aspect even: the dimension of the text-block and the line number is the same and both scripts are in the naskh style. But even an eye unaccustomed to Arabic script will feel the presence of crude restoration work. The paleographic expertise can be automated; the magnitude of a 2D Fast Fourier Transform of the pages will reveal two main differences. The axis-angle of the cross-like spectra are different, which means that the vertical strokes of the replacement script are more slanted than the original ones (Fig. 1c has a nearly horizontal axis, Fig. 1d a more oblique one; the vertical axis comes from the regular spacing of the 13 lines). The four bright spots away from the center of Fig. 1d suggest a stronger linearity of the replacement script, whereas the dispersion in Fig. 1c points to a rounder and more irregular handwriting in the original script). But there is an even more powerful distinguishing feature: two allographs of the kaf letter are used by the replacement, the oblique one (marked by a circle, 13 times in Fig. 1f ) and the long one (marked with a disc, 7 times in Fig. 1f ), whereas only the oblique appears in the original (19 times in Fig. 1e). It is also remarkable that most (4) come in the context of the same word (kana “to be”). Defined by quantities and contexts never obvious to the eye, the allographs offer a reliable security filigree for testing the integrity of the manuscript. Applications Figs. 2a and 2b are excellent examples of the principal application of allometry: biometrics, or the spatiotemporal identification of scripts and their attribution to individuals or traditions (a subject of interest to police and intelligence agencies, the judicial system, the art market, historians, et al.). The examples presented here are written by two calligraphers for a competition and transcribe the same text [2, p. 31–32]; however, the number of long kaf -s is 6 and 23, respectively. Thus the allographs of the elongated type clearly play a very different role as “visual attractors” from one individual to 328 TUGboat, Volume 24 (2003), No. 3—Proceedings of EuroTEX 2003 Allographic Biometrics and Behavior Synthesis R e p la c e m e n t s c ri p t D O C U M E N T Figure 1 S H A P E A N A LY S IS A L L O G R A P H A N A LY S IS (b) F o u ri e r tr a n s fo rm