Overview - ICDAR 2019 Robust Reading Challenge on Scanned Receipts OCR and Information Extraction
Scanned receipts OCR is a process of recognizing text from scanned structured and semi-structured receipts, and invoices in general. On the other hand, extracting key texts from receipts and invoices and save the texts to structured documents can serve many applications and services, such as efficient archiving, fast indexing and document analytics. Scanned receipts OCR and information extraction (SROIE) play critical roles in streamlining document-intensive processes and office automation in many financial, accounting and taxation areas. However, SROIE also faces big challenges. With performance greatly boosted by recent breakthroughs in deep learning technologies in terms of accuracy and processing speed, OCR is becoming mature for many practical tasks (such as name card recognition, license plate recognition and hand-written text recognition). However, receipts OCR has much higher accuracy requirements than the general OCR tasks for many commercial applications. And SROIE becomes more challenging when the scanned receipts have low quality. Therefore, in the existing SROIE systems, human resources are still heavily used in SROIE. There is an urgent need to research and develop fast, efficient and robust SROIE systems to reduce and even eliminate manual work.
With the trends of OCR systems going to be more intelligent and document analytics, SROIE holds unprecedented potentials and opportunities, which attracted huge interests from big companies, such as Google, Baidu and Alibaba. Surprisingly, there are little research works published in the topic of SROIE. While robust reading, document layout analysis and named entity recognition are relevant to the SROIE, none of the existing research and past ICDAR competitions fully address the problems faced by SROIE [1,2,3].
In recognition of the above challenges, importance and huge commercial potentials of SROIE, we propose the ICDAR 2019 competition on SROIE, aiming to draw attention from the community and promote research and development efforts on SROIE. The proposed competition could be of interests to the ICDAR community from several aspects:
- To support the competition, a large-scale and well-annotated invoice datasets are provided, which is crucial to the success of deep learning based OCR systems. While many datasets have been collected for OCR research and competitions, to the best of our knowledge, there is no publicly available receipt dataset. Compared to the existing ICDAR and other OCR datasets, the new dataset has some special features and challenges, e.g., some receipts having poor paper quality, poor ink and printing quality; low resolution scanner and scanning distortion; folded invoices; too many unneeded interfering texts in complex layouts; long texts and small font sizes. To address the potential privacy issue, some sensitive fields (such as name, address and contact number etc) of the receipts are blurred. The datasets can be an excellent complement to the existing ICDAR and other OCR datasets.
- Two specific tasks are proposed: receipt OCR and key information extraction. Compared to the other widely studied OCR tasks for ICDAR, receipt OCR (including text detection and recognition) is a much less studied problem and has some unique challenges. On the other hand, research works on extraction of key information from receipts have been rarely published. The traditional approaches used in the named entity recognition research are not directly applicable to the second task of SROIE.
- Comprehensive evaluation method is developed for the two competition tasks. In combination with the new receipt datasets, it enables wide development, evaluation and enhancement of OCR and information extraction technologies for SROIE. It will help attract interests on SROIE, inspire new insights, ideas and approaches.
. D. Karatzas, F. Shafait, S. Uchida, M. Iwamura, L. Gomez, S. Robles, J. Mas, D. Fernandez, J. Almazan, L.P. de las Heras: ICDAR 2013 Robust Reading Competition. ICDAR, 2013.
. D. Karatzas, L. Gomez-Bigorda, A. Nicolaou, D. Ghosh , A. Bagdanov, M. Iwamura, J. Matas, L. Neumann, VR. Chandrasekhar, S. Lu, F. Shafait, S. Uchida, E. Valveny: ICDAR 2015 robust reading competition. ICDAR, 2015.
. Everingham, M. and Eslami, S. M. A. and Van Gool, L. and Williams, C. K. I. and Winn, J. and Zisserman, A.: The Pascal Visual Object Classes Challenge: A Retrospective. IJCV, 2015.
. D. Karatzas, L. Rushinol, The Robust Reading Competition Annotation and Evaluation Platform.
Registration open: February 10 – March 31, 2019
Training/validation dataset available: March 1, 2019
Task 1&2 submission open: April 15, 2019
Task 1&2 submission deadline: April 22, 2019
Task 3 submission open: April 23, 2019
Task 3 submission Deadline: May 5, 2019
Note on the registration for the SROIE challenge:
There is no need to register explicitly for the SROIE challenge. As long as you are registered to the RRC portal you will be able to submit your results when the submission is open.
Note on the granulairty level of recognition outcomes for Task2:
Dear All, we received a few quiries on the above issue. We would like to clarify that for Task2, you should submit all the recognized words, which are split from the detected sentences. For example, for a sentence "Payment Date: 2018/09/19", you should submit three words "Payment", "Date:" and "2018/09/19". The chinese characters if any need to be ignored in the submission.
Note on the extension of Task 3 submission deadline:
Dear All, following many requests from Task 3 participants, the Task 3 submission deadline is extended to 5th May, 2019.