Allograph-Based Categorization of Handwriting Styles

Abstract: The aim of this work is to aid the development of online handwriting recognition software by using a linguistic approach to the problem of categorizing handwriting styles. This is done by exploring the wide variety of styles among users of the Latin alphabet, trying to explain the variety and finding ways of categorizing the styles. The variety is due to the many variables open to the writer and the wide range of these variables. A generative model for the production of hndwriting is proposed. Different ways of categorizing styles along the lines of the variables are explored, and one of them, allographic variation, i.e. different models of the same letter, is selected or the study of a database of online samples consisting of handwritten block letters from six European countries. Not only the shapes of letters are examined but also the number and sequence of the constituting penstrokes, which is especially interesting from an online recognition point of view. It is found that there are different 'alphabets' or categories of allographs, as well as geographical differences in the adherence to the alphabets.