21th AIAI 2025, 26 - 29 June 2025, Limassol, Cyprus

An Integrated Convolutional & Transformer Architecture for Word-Based Handwriter Identification

Majithia Aditya, Pedersen Arthur Paul, Grossberg Michael

Abstract:

  This paper focuses on the problem of developing automated methods for writer identification from handwriting. It introduces CONVOLUTIONAL TRANSFORMER ENCODER (CTE), a novel architecture designed for determining authorship from handwritten text, be it modern or historical handwriting. CTE is the first of its kind designed to operate on historical handwritten fragments, setting it apart from existing methods that rely on entire document pages. The implementation of CTE across multiple handwriting datasets is shown to enjoy superior performance over state-of-the-art lexical-segmentation methods for modern handwriting datasets and competitive performance with folio-directed methods on historic handwriting datasets. CTE tracks subtle features and cues from handwriting strokes to distinguish authorship.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.