The NLPR-CASIA Wins Multiple Champions in Competitions of ICDAR2017
Nov 23, 2017Author:
PrintText Size A A

The 14th International Conference on Document Analysis and Recognition (ICDAR2017) was held in Kyoto, Japan from November 9th to November 15th, 2017. More than 500 people from academia and industry participated in this conference at such best season.

A group from the National Laboratory of Pattern Recognition, Institute of Automation of the Chinese Academy of Sciences (NLPR-CASIA), called Pattern Analysis and Learning Group (PAL Group), led by Prof. Chenglin Liu, participated in the conference this year. They have seven papers selected by ICDAR2017, including five orals and two posters. In these papers they shared their research progress and achievements in text detection, character recognition, layout analysis and writing style adaptation.

The results of the algorithm competitions of ICDAR and the related technologies have drawn highly attention. This year, the PAL group participated in 6competitions and won the first place in 8tasks, the second place in 2tasks. Specifically, in the Competition on Page Object Detection, they won the first place in all 4 tasks (formula detection, table detection, figure detection and page object detection). In the Competition on Layout Analysis for Challenging Medieval Manuscripts, they won the first place in the layout analysis task. In the Competition on Arabic Text Detection and Recognition in Multi-resolution Video Frames, they won the first place in the video textile recognition task. In the Competition on Reading Chinese Text in the Wild, they won the first place in the End-to-End Recognition task, and the second place in the text detection task. In addition, in the Competition on Robust Reading Challenge in Omnidirectional Video, they won the first place in the text localization task.

The PAL group has been making a long-term and in-depth study on document image processing, layout analysis, text detection, character recognition, text line recognition, context modeling and the related theories of pattern recognition and machine learning. Their algorithms and techniques have been transferred to real applications of bank check recognition, web document analysis, and handwritten document digitization. In order to promote academic research and technology development, the group released a large database of Chinese handwritten characters and documents written by more than 1,000 people, which has been licensed to hundreds of research groups. The PAL group also organized three competitions on Chinese handwriting recognition in 2010, 2011 and 2013. This year, collaborating with the University of La Rochelle (France) and Samsung Research Institute China, the PAL group organized a competition on multilingual scene text detection and script identification. In this competition, they released a multilingual scene text image database containing 18,000 images of 9 languages.

The conference ICDAR was started from 1991 in Saint Malo, France and is taken place biennially. It is a very successful and flagship conference series in the field of document analysis and recognition, and is one of biggest and premier international conferences sponsored by the IAPR (International Association for Patter Recognition). In recent years, Chinese scholars have produced growing impacts in the field of document analysis and recognition. In ICDAR2017, Chinese scholars contributed the highest number of submissions with their tireless efforts. In addition to the Institute of Automation of Chinese Academy of Sciences, other institutions such as Peking University, Tsinghua University, South China University of Technology, Huazhong University of Science and Technology, University of Science and Technology Beijing, Samsung Research Institute China, Tencent and Baidu, also organized or participated in ICDAR competitions actively. Professor Xiang Bai, Huazhong University of Science and Technology, was invited to give a keynote speech in ICDAR2017.He became the first keynote speaker from Chinese academia in ICDAR history.