Convolutional Swin Encoder
A Unified Deep Learning Approach to Writer Attribute Prediction
DOI:
https://doi.org/10.32473/flairs.38.1.138949Keywords:
Authorship Attribution, Handwriting Analysis, Swin Transformers, multiple task learningAbstract
This paper focuses on developing a deep learning architecture capable of identifying writers' attributes from their handwriting. It introduces Convolutional Swin Encoder (CSE), a novel architecture combining Visual Geometry Group Network (VGGNet) and Swin Transformer blocks. CSE is designed to handle multi-label classification using images of individual handwritten words. As a unified encoder, it can predict writers' attributes such as identity, gender, age, and handedness. Using a word-level segmentation approach, CSE achieves competitive performance compared to page-level methods, which typically rely on separate classifiers instead of a unified one.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Aditya Majithia, Arthur Paul Pedersen, Michael Grossberg

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.