Call for papers

Pattern recognition is the foundation of modern document image understanding, supporting progress in document classification, handwritten text recognition, DocVQA, and multimodal modeling of text, layout, and visual structure. As these systems are increasingly deployed in real-world, high-stakes environments, they must meet new expectations for trust, safety, and regulatory compliance. Current document image models are required to support machine unlearning, to resist imperceptible document forgeries and membership inference attacks, to preserve the privacy of sensitive handwritten or scanned data, and to offer transparent and interpretable decisions. These demands raise fundamental research questions on how models memorise and forget document-specific features, how handwritten text recognition can remain robust under regime changes, and how multimodal document representations should be evaluated from ethical and trustworthiness perspectives. 

The workshop invites contributions that advance this emerging area. It specifically welcomes contributions on topics including (but not limited to):

Machine unlearning in document AI

– unlearning for document image classification

– unlearning for handwritten text recognition

– unlearning for document visual question answering

– unlearning for multimodal document understanding

Robustness in document image recognition systems

– adversarial attacks & defenses

– robustness to data noise, layout variation, and handwriting styles

Privacy in document image understanding

– differential privacy for document image datasets

– membership inference and training data leakage prevention

– privacy-preserving OCR and document representation learning

Explainability and interpretability

– explaining document model decisions at pixel, region, and semantic levels

– visualization tools for understanding features learned from document images

Evaluation, benchmarks, and best practices

– metrics and datasets for assessing unlearning, privacy, robustness, and explainability

– standardized evaluation pipelines for trustworthy document intelligence

Applications and case studies

– legal, financial, medical, and government documents

– lifelong learning and compliance-driven document AI

Submission Guidelines

Paper Format and Templates

Accepted workshop papers will be published in the LNCS proceedings. Authors must use the LNCS template provided on the ICPR 2026 website (DOC Template, LaTeX Template). Papers may be up to 15 pages (including references). Only full papers (more than 6 pages) will be included in the proceedings.

All submissions must fully comply with the LNCS formatting guidelines. Papers that do not follow the format may be rejected without review.

Submitted papers must contain previously unpublished, original work of high quality. Dual submission to other venues with proceedings is not allowed.

Anonymized Submission

This workshop uses a double-blind peer-review process, meaning that author names, affiliations, and identifiable references must be removed.

Improperly anonymized submissions may be desk rejected.

Publication

Accepted papers will be published in the ICPR Workshops Proceedings within the Lecture Notes in Computer Science (LNCS), Springer series. All accepted papers must be presented at the workshop.

Submission Site

Please submit your paper through CMT: https://cmt3.research.microsoft.com/TrustDOC2026/

Organizing Committee

• Francesco Pinto, Nathalie Rauschmayr, Florian Tramèr, Philip Torr, and Federico Tombari. "Extracting training data from document-based VQA models." In International Conference on Machine Learning (ICML), 2024.
• Rubèn Tito, Khanh Nguyen, Marlon Tobaben, Raouf Kerkouche, Mohamed Ali Souibgui, Kangsoo Jung, Joonas Jälkö et al. "Privacy-aware document visual question answering." In International Conference on Document Analysis and Recognition (ICDAR), 2024.
• Lei Kang, Mohamed Ali Souibgui, Fei Yang, Lluis Gomez, Ernest Valveny, and Dimosthenis Karatzas. "Machine unlearning for document classification." In International Conference on Document Analysis and Recognition (ICDAR), 2024.
• Mohamed Ali Souibgui, Changkyu Choi, Andrey Barsky, Kangsoo Jung, Ernest Valveny, and Dimosthenis Karatzas. "DocVXQA: Context-Aware Visual Explanations for Document Question Answering." In International Conference on Machine Learning (ICML), 2025.
• Khanh Nguyen, Raouf Kerkouche, Mario Fritz, and Dimosthenis Karatzas. "DocMIA: Document-Level Membership Inference Attacks against DocVQA Models." In International Conference on Learning Representations (ICLR), 2025.
• Marco Pintore, Maura Pintor, Dimosthenis Karatzas, and Battista Biggio. "Counterfeit Answers: Adversarial Forgery against OCR-Free Document Visual Question Answering." arXiv preprint arXiv:2512.04554 (2025).
• Dong Nguyen Tien, and Dung D. Le. "Robustness Evaluation of OCR-based Visual Document Understanding under Multi-Modal Adversarial Attacks." arXiv preprint arXiv:2506.16407 (2025).
• Lei Kang, Xuanshuo Fu, Lluis Gomez, Alicia Fornés, Ernest Valveny, and Dimosthenis Karatzas. "Preserving Privacy Without Compromising Accuracy: Machine Unlearning for Handwritten Text Recognition." In Pattern Recognition 172, 112411, 2026.

The Microsoft CMT service was used for managing the peer-reviewing process for this conference. This service was provided for free by Microsoft and they bore all expenses, including costs for Azure cloud services as well as for software development and support.