MMK @ ICDAR2023

To kick off a busy conference autumn season, members from the MMK project team at SBB participated in the International Conference on Document Analysis and Recognition (ICDAR2023) conference, held from August 21-26, 2023 in San José, California, USA. As the main event of the International Association for Pattern Recognition (IAPR), ICDAR was already held for the 17th time in 2023 and has established itself as the premier international scientific conference in the field of document analysis and recognition, with an average of 500 international attendees from research and also industry.

The ICDAR2023 conference program featured a large variety of talks, competitions, panels, tutorials and workshops with three keynotes. An overview of the most common submission topics can be found below, indicating a strong interest in historical and handwritten documents.

It was particularly nice to also meet other librarians at ICDAR, like Dr Adi Keinan-Schoonbaert, Digital Curator for Asian and African Collections at the British Library, who wrote a nice blog on her participation, and Dr Mirjam Cuper, Data scientist at the National Library of the Netherlands, who presented a wonderful poster and paper on confidence scores as proxy for OCR quality. 

Since 2011, one of the satellite workshops attached to the ICDAR conference is HIP, the International Workshop on Historical Image Processing, which puts a particular focus on historical documents and use cases and applications from digital libraries and the digital humanities. For the second time after 2021, HIP was jointly organised by the Berlin State Library and the Pattern Recognition and Image Analysis (PRImA) Research Lab at the University of Salford, Greater Manchester. 

HIP’23 featured a full day program with peer-reviewed paper presentations and a half day workshop excursion to the Computer History Museum. HIP’23 was again a highly successful workshop with strong participation, attracting 32 submissions (18 accepted) from authors representing 19 different countries. The full workshop proceedings are already published in the ACM Digital Library.

The MMK team had a total of three contributions accepted for the HIP’23 workshop:

1. Kai Labusch, Clemens Neudecker. Gauging the Limitations of Natural Language Supervised Text-Image Metrics Learning by Iconclass Visual Concepts
   [slides] [paper] [code]
2. Vahid Rezanezhad, Konstantin Baierer, Mike Gerber, Kai Labusch and Clemens Neudecker. Document Layout Analysis with Deep Learning and Heuristics
   [slides] [paper] [code]
3. Vahid Rezanezhad, Konstantin Baierer and Clemens Neudecker. A hybrid CNN-Transformer model for Historical Document Image Binarization
   [slides] [paper] [code]

Looking back, it was particularly nice to meet many international colleagues again after the pandemic prevented most from attending the last ICDAR2021 in Lausanne, Switzerland, in person.

Looking forward, one of the key developments is that due to its ongoing success, the HIP workshop will become an officially endorsed IAPR workshop within the TC-11, meaning that future calls for papers can be advertised a lot earlier (but still allow submissions after the ICDAR deadline), and that the workshop proceedings will move from ACM to Springer, who have in recent years followed the route towards open access for computer science conference proceedings.