Script independent offline writer identification using handwriting style

A common practice in writer identification is to classify handwriting style using character shapes (allographs). Allograph codebooks are used for this purpose. Generating codebooks relies on the availability of template characters that are representative for the style. Due to handwriting individuality, limited reference data, and unequal consistency of writers in using allographs, complete coverage of all possible allographs is complex. Allograph segmentation is also a paradox to the solution. This paper proposes an alternate approach to identify writers using handwriting style. Handwriting is taken as consist of n-branching ink fragments originating from a central point. Probability distribution functions (PDFs) of these branching structures´ orientation (branchlet distributions) as features overcomes allographic constraints. Promising results are obtained using this approach.