Interpretable Writer Recognition via Vectors of Locally Aggregated Characters

Raven T, Christlein V, Fink GA (2025)

Publication Type: Conference contribution

Publication year: 2025

Journal

Lecture Notes in Computer Science Springer Verlag

Series: Lecture Notes in Computer Science

Pages Range: 429-445

Conference Proceedings Title: Document Analysis and Recognition – ICDAR 2025

Event location: Wuhan CN

ISBN: 9783032046260

DOI: 10.1007/978-3-032-04627-7_25

Abstract

Writer recognition involves analyzing handwritten documents with respect to the identity of the writer. Although automated methods can achieve strong benchmark results, their lack of interpretability limits practical adoption, particularly in settings where trust and verifiability are critical.


To address this challenge, we propose a novel framework grounded in character-level analysis. At its core is Vectors of Locally Aggregated Characters (VLAC), a feature aggregation method that fuses the aligned outputs of a feature extractor and a feature annotator network. By aggregating local features on a per-character basis, VLAC provides the backbone for computing verifiable character-wise distances, thereby enhancing interpretability and trustworthiness.


We extensively evaluate the proposed framework on two contemporary datasets (CVL and IAM) – achieving new state-of-the-art retrieval results on CVL – as well as a historical dataset (Hist-WI). Our method does not only perform well, but also facilitates interpretable insights into the decision process, paving the way for broader acceptance in practical and forensic applications.

Authors with CRIS profile

Vincent Christlein Lehrstuhl für Informatik 5 (Mustererkennung)

Involved external institutions

Technische Universität Dortmund DE Germany (DE)

How to cite

APA:

Raven, T., Christlein, V., & Fink, G.A. (2025). Interpretable Writer Recognition via Vectors of Locally Aggregated Characters. In Document Analysis and Recognition – ICDAR 2025 (pp. 429-445). Wuhan, CN.

MLA:

Raven, Tim, Vincent Christlein, and Gernot A. Fink. "Interpretable Writer Recognition via Vectors of Locally Aggregated Characters." Proceedings of the Document Analysis and Recognition – ICDAR 2025, Wuhan 2025. 429-445.

BibTeX: Download