Recently, the 25th International Conference on Pattern and Recognition (ICPR) was held online. The results of the third Competition on Text Detection and Recognition in Arabic News Video Frames (AcTiVComp) were announced, and Ph.D. candidate Yan Ruijie and master’s student Xiao Shanyu, advised by Associate Professor Peng Liangrui from the Department of Electronic Engineering at Tsinghua University, were champions in both tasks of text detection and text recognition.
AcTiVComp was organized by the HES-SO University of Applied Sciences and Arts of Western Switzerland, the University of Fribourg in Switzerland, and the University of Sousse in Tunisia. Participants came from China, Switzerland, Malaysia, Norway, India, Pakistan, and other countries.
Multilingual OCR (Optical Character Recognition) technologies including Arabic OCR are crucial in global information communication and utilization, which are also cutting-edge research topics in the machine learning and artificial intelligence research fields. Advised by Peng Liangrui, the graduate students participating in the AcTiVComp include Yan Ruijie, Xiao Shanyu, Yao Gang and Shi Haodong. The text detection algorithm was mainly developed by Xiao Shanyu, and had novel contributions in deep learning model architecture and multi-task supervised learning; the text recognition algorithm was mainly developed by Yan Ruijie, and showed breakthroughs in efficient feature representation learning and transfer learning. Peng Liangrui’s research group has previously won both tasks of text detection and recognition in the AcTivComp hosted by ICDAR (International Conference on Document Analysis and Recognition) in 2017. Compared with the algorithms in 2017, the newest algorithms have achieved significant performance improvements in both text detection and text recognition.
The related research work was supported by the National Key R&D Program of China, the Institute for Guo Qiang at Tsinghua University, and the Beijing National Research Center for Information Science and Technology.
Editors: Li Han, John Olbrich