IAPR TC11 Newsletter 2019 11
Click on the buttons below to view sections of the newsletter.
- Message from the Editor
- Dates and Deadlines
- Upcoming Conferences and Events
- DAS 2020: Call for Papers *(repost)*
- ICFHR 2020: Call for Papers *(repost)*
- ICFHR 2020: Call for Competitions
- IJDAR: Special Issue (Vol. 22, Issue 3) *(repost)*
- TC11 Datasets Repository *(repost)*
- PhD Subject: use of graph attention networks for named entity extraction in scanned documents
- Student Industrial Internship Opportunities (IAPR) *(repost)*
Before the end of the year, there are several submission deadlines coming up. Please consider submitting your paper abstracts for DAS 2020 until December 13th and full papers for ICPRAI 2020 until December 15th (extended). Furthermore, competition proposals for ICFHR 2020 can be submitted until December 13th.
Looking for a PhD position? Then please have a look at the open position in Nancy, France, in the career section of this newsletter.
Andreas Fischer, TC11 Communications Officer
( email@example.com )
- Dec. 13: Paper abstract submission DAS 2020
- Dec. 13: Competition proposal submission ICFHR2020
- Dec. 15: Paper submission ICPRAI 2020
- March 1: Paper submission ICFHR 2020
Upcoming Conferences and Events
- ICPRAI 2020. Zhongshan, China (May 12-15, 2020)
- DAS 2020. Wuhan, China (May 17-20, 2020)
- ICFHR 2020. Dortmund, Germany (September 8-10, 2020)
2021 and Later
- ICDAR 2021. Lausanne, Switzerland (September 5-10, 2021)
- ICFHR 2022. Hyderabad, India (December, 2022)
DAS 2020: Call for Papers (repost)
PDF Version of this Call: das2020_call_for_papers.pdf
Regular paper submission Dec 13, 2019 Abstract due Dec 20, 2019 Full Papers due Feb 15, 2020 Notification of acceptance Tutorial proposals Jan 10, 2020 Proposals due Short paper submission Mar 10, 2020 Papers due Mar 24, 2020 Notification of acceptance
DAS 2020 is the 14th edition of the workshop focusing on system-level issues and approaches in document analysis and recognition. The workshop comprises invited speaker presentations, oral, poster, tutorial and demo sessions as well as working group discussions. Proceedings (full papers) will be published by Springer.
Topics & Technologies:
- Document analysis systems
- Document understanding
- Layout analysis
- Deep learning for document analysis systems
- Document analysis for digital humanities
- Document analysis for libraries and archives
- Document analysis for the internet
- Document analysis for mobile devices
- Camera-based document analysis
- Document datasets
- Document retrieval
- Information extraction from document images
- Graphics recognition
- Table and form processing
- Mathematical expression recognition
- Document authentication
- Document image watermarking
- Forensic document analysis
- Historical document analysis
- Multilingual document analysis
- Multimedia document analysis
- Pen-based input and its analysis
- Authoring, annotation, and presentation systems
- Performance evaluation
DAS2020 will include both long and short papers, posters and demonstrations of working or prototype systems. All submissions will undergo a rigorous review process with a minimum of 3 reviews considering the originality of work, the quality of research or analysis of experience, the relevance to document analysis systems, and quality of presentation of ideas.
Full papers should describe complete works of original research. Authors are invited to submit original, unpublished research papers that are not being considered in another forum, up to 6 pages length.
Short papers provide an opportunity to report on research in progress, to present novel positions on document analysis systems. Authors may submit short papers (up to 2 pages in length). Short papers will also undergo review and will appear in an extra booklet, not in the official DAS2020 proceedings.
DAS2020 seeks public demonstrations of novel systems, to be presented to the workshop as a whole during breaks between paper sessions. Prospective demonstrators should submit a two-page summary of the system to be demonstrated. Demonstration proposals will undergo review and summaries of accepted demonstrations will appear in the extra DAS2020 booklet, as in the case of short papers.
Cheng-Lin Liu, Shijian Lu, and Jean-Marc Ogier, DAS 2020 General Chairs
( firstname.lastname@example.org )
ICFHR 2020: Call for Papers (repost)
PDF Version of this Call: ICFHR2020-1stCfP.pdf
Mar 01, 2020 Paper Submission Jun 08, 2020 Author Notifications Jul 12, 2020 Camera-Ready Papers Due
The International Conference on Frontiers of Handwriting Recognition (ICFHR) is the premier scientific venue in the field of handwriting recognition. This conference brings together international experts from academia and industry to share their experiences and to promote research and development in all aspects of handwriting recognition and applications.
Topics of interest to the conference include, but are not limited to:
- Handwriting Recognition
- Cursive Script Recognition
- Symbol, Equation, Sketch and Drawing Recognition
- Word Spotting
- Handwritten Document Image Processing
- Layout Analysis and Understanding
- Language Models in Handwriting Recognition
- Web-Based Applications
- Handwritten Databases and Digital Libraries
- Information Extraction & Retrieval
- Form Processing
- Bank-Check Processing
- Historical Document Processing
- Forensic Studies and Security Issues
- Writer Verification and Identification
- Performance Enhancement and System Evaluation
- Electronic Ink and Pen-Based Systems
- Other Offline and Online Applications
Authors are invited to submit full-length papers of not more than six (6) pages. Papers must describe original work. Paper reviews will be double blind. Instructions for paper submission will be available on the ICFHR 2020 web site (http://icfhr2020.org).
Gernot A. Fink and Lambert Schomaker, ICFHR 2020 General Chairs
( email@example.com )
ICFHR 2020: Call for Competitions
PDF Version of this Call: CallForCompetition_ICFHR2020_v1.pdf
Dec 13, 2019 Submission of competition proposals Dec 20, 2019 Notification of acceptance Dec 24, 2019 Competitions open to participants Jun 14, 2020 Submission of competition paper Jun 26, 2020 Notification of acceptance
The ICFHR 2020 Organizing Committee invites proposals for competitions addressing current research challenges in handwriting recognition and related areas of pattern recognition and vision. Competitions should aim at evaluating the performance of individual methods and systems that fall within the context of the conference’s call for papers. Competitions should also participate in the sharing and standardization of handwriting recognition processes.
Proposals must contain the following information:
- A brief description of the competition, including which is the particular task under evaluation and why this competition could be of interest to the ICFHR community. If a similar competition has been held in a previous edition of ICFHR, the novel additions of the proposed one must be clearly explained.
- A draft of the outline of the competition describing which data is planned to be used, how will the submitted methods be evaluated and which performance measures will be used. Proposals should state explicitly whether data and annotations are already available or otherwise the timeframe to prepare them.
- How datasets and evaluation scripts will be made publicly available.
- The list of rules or best practices which will be imposed to participants (allowing multiple submissions or not, allowing external data or not, public ranking or not, etc.). Each competition relies on the academic integrity of the competitors.
- The names, contact information, and brief CVs of the competition organizers, outlining previous experience in performance evaluation and/or organizing competitions.
Proposals should be submitted by electronic mail to the ICFHR2020 Competitions Chairs:
Harold Mouchère and Dimosthenis Karatzas (firstname.lastname@example.org)
The following rules will apply to the accepted competitions:
- All competitions must run well in advance of the conference.
- Datasets used in the competitions must be made available through the TC10/TC11 Online Resources website after the end of the competitions (or equivalent arrangement by approval of the competition chairs).
- Evaluation methodologies and metrics used must be described in detail so that results can be replicated later.
- The name of the competition must be standardized by starting with “ICFHR2020” e.g. “ICFHR2020 Competition on …”;
- Reports (full papers) on each competition will be reviewed and, if accepted (the competition run according to plan and is appropriately described), will be published in the ICFHR2020 conference proceedings.
- The results of accepted competitions will be announced during a dedicated session of the conference. Competition organizers are expected to participate to this dedicated session during ICFHR 2020. Participants should be encouraged to present their algorithms (if not already done) in a conference paper at ICFHR 2020.
- Participants should be encouraged to share their algorithms and code using one of the open-source licenses
- Competitions must have sufficient number of participants (about 5, depending of the originality of the topic) to be able to draw meaningful conclusions.
- Each competition will be allocated some time in parallel to the poster session, during which they can run a mini-workshop (e.g. presenting results and conclusions, inviting winning methods to present their methods, lessons learnt, etc).
- Each competition accepted to present results in the conference, should prepare a poster presentation and present it during ICFHR 2020.
Harold Mouchère and Dimosthenis Karatzas, ICFHR 2020 Competition Chairs
( email@example.com )
IJDAR: Special Issue (Vol. 22, Issue 3) (repost)
The IJDAR Special Issue on Advanced Topics in Document Analysis and Recognition has been released. The articles have been presented orally at ICDAR2019 during the journal track session. Click on the links below to go directly to the Springer Link page for each article.
Table of Contents
- Are 2D-LSTM really dead for offline text recognition? Bastien Moysset & Ronaldo Messina
- Boosting scene character recognition by learning canonical forms of glyphs. Yizhi Wang, Zhouhui Lian, Yingmin Tang & Jianguo Xiao
- Generalized framework for summarization of fixed-camera lecture videos by detecting and binarizing handwritten content. Bhargava Urala Kota, Kenny Davila, Alexander Stone, Srirangaraj Setlur & Venu Govindaraju
- Dynamic temporal residual network for sequence modeling. Ruijie Yan, Liangrui Peng, Shanyu Xiao, Michael T. Johnson & Shengjin Wang
- A comparison of local features for camera-based document image retrieval and spotting. Quoc Bao Dang, Mickal Coustaty, Muhammad Muzzamil Luqman & Jean-Marc Ogier
- Comic MTL: optimized multi-task learning for comic book image analysis. Nhu-Van Nguyen, Christophe Rigaud & Jean-Christophe Burie
- A two-stage method for text line detection in historical documents. Tobias Grüning, Gundram Leifert, Tobias Strauß, Johannes Michael & Roger Labahn
- On optimal stopping strategies for text recognition in a video stream as an application of a monotone sequential decision model. Konstantin Bulatov, Nikita Razumnyi & Vladimir V. Arlazarov
- An anchor-free region proposal network for Faster R-CNN-based text detection approaches. Zhuoyao Zhong, Lei Sun & Qiang Huo
- Handwritten Arabic text recognition using multi-stage sub-core-shape HMMs. Irfan Ahmad & Gernot Fink
- Coarse-to-fine document localization in natural scene image with regional attention and recursive corner refinement. Anna Zhu, Chen Zhang, Zhi Li & Shengwu Xiong
TC11 Datasets Repository (repost)
TC11 maintains a collection of datasets that can be found online in the TC11 Datasets Repository.
If you have new datasets (e.g., from competitions) that you wish to share with the research community, please use the online upload form. For questions and support, please contact the TC11 Dataset Curator (contact information is below).
Joseph Chazalon (TC11 Dataset Curator)
( firstname.lastname@example.org )
PhD Subject: use of graph attention networks for named entity extraction in scanned documents
Contact: Abdel Belaïd, LORIA lab, READ group, Nancy, France, email@example.com
Since the entry into force of the RGPD (General Regulation on Data Protection) on May 25, 2018, European citizens have the legal basis necessary to control and reclaim their personal data. It is a question of providing the technical means that are simple and accessible to as many people as possible to administer their data, to share it, while allowing the organizations that use it to comply with the legislation in force. For this, we want to use massively artificial intelligence to provide a disruptive approach to the management of user data and especially their documents. It is a question of proposing a totally automatic management of the documents of the users, as well at the level of the classification as of the exploitation of the information which they contain. This information (metadata) will then be integrated into an ontology of personal data in order to interrogate these data in a structured way but above all to propose automated reasoning algorithms.
A first solution has been proposed for the extraction of named entities in personal documents of the “Invoice” type. Documents are scanned, and named entities are extracted from images after Optical Character Recognition (OCR) reading. These entities correspond to the information relating to the issuing company, the delivery and billing addresses, the products ordered as well as the different prices and totals. A deep learning technique based on convolutional network graphs (GCN) has been implemented for the extraction of these entities . It consists in taking advantage of the contextual links in the neighborhoods of the words of the document. Each node of the graph represents a word of the document and its relations connect it to its four closest neighbors in the four cardinal directions. For the purposes of convolution, the graph in Euclidean space is replaced by the adjacency matrix describing the words in the spectral domain. Thus, the model learns to individually label the words of the document based on their neighborhood links. In this project, the READ team seeks to extend the extraction method to other types of documents in order to test its ability to extract named entities of different types and in different types of media. We want to test and compare graph attention networks (GAN) of [4, 5, 6, 7, 8] and see which is best suited to the problem. The aim is to better specialize neighborhood graphs to better represent semi-structured information, by retaining the nodes around the headers or information indicators, because it will be tedious to use all the words in the document. This was the case for invoices that are very structured, which is not the case for all documents.
The implementation will be done in Python using the Keras API based on the Tensorflow library.
 Devashish Lohani, Abdel Belaid, Yolande Belaid. An Invoice Reading System Using a Graph Convolutional Network International Workshop on Robust Reading, Dec 2018, PERTH, Australia. 2018.
 Schlichtkrull, M., T. N. Kipf, P. Bloem, R. van den Berg, I. Titov, and M. Welling 2018. Modeling Relational Data with Graph Convolutional Networks. In Lecture Notes in Computer Science (including subseries Reading, Notes in Artificial Intelligence and Reading Notes in Bioinformatics), volume 10843 LNCS, pp. 593-607.
 Jianan Li, Jimei Yang, Aaron Hertzmann, Zhang Jianming, Tingfa Xu, Layoutgan: Generating Graphic Layouts with Wireframe Discriminators, ICLR 2019, pp. 1-16.
 Velickovic, P., G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio 2017. Graph Attention Networks.
 Gong, L., and Q. Cheng, 2018. Adaptive Edge Features Guided Graph Attention Networks.
 Zhang, J., X. Shi, J. Xie, H. Ma, I. King, and D.-Y. Yeung 2018. GaAn: Gated Attention Networks for Learning on Large and Spatiotemporal Graphs.
 Monti, F., O. Shchur, A. Bojchevski, O. Litany, S. Günnemann, and M. M. Bronstein 2018. Dual-Primal Graph Convolutional Networks. Pp. 1 to 11.
 John Boaz Lee, Ryan A. Rossi, Kim Sungchul, Nesreen Kamel Ahmed, Eunyee Koh, 2018. Attention Models in Graphs: A Survey. 0 (1).
Student Industrial Internship Opportunities (IAPR) (repost)
IAPR’s Industrial Liaison Committee is pleased to announce the opening of its Company Internship Brokerage List.
The web page lists internship opportunities for students at different levels of education and specialism. We expect many additional internship opportunities to be listed here as the community becomes more aware of the site.
IAPR Company Internship Brokerage List:
Bob Fisher, Chair, IAPR Industrial Liason Committee
( firstname.lastname@example.org )
Call for Contributions: To contribute news items, please send a short email to the editor, Andreas Fischer (). Contributions might include conference and workshop announcements/updates/reports, career opportunities, book reviews, or anything else of interest to the TC-11 community.
Subscription: This newsletter is sent to subscribers of the IAPR TC11 mailing list. To join the TC-11 mailing list, please click on this link: Join the TC-11 Mailing List. To manage your subscription, please visit the mailing list homepage: TC-11 Mailing List Homepage.