I²R and NUS triumph with winning formula at DIBCO 2009

Printer-friendly versionPDF version

Dr Lu Shijian from A*STAR Institute for Infocomm Research (I2R) and Professor Tan Chew Lim from National University of Singapore’s School of Computing have beaten thirty-five other competitors to clinch the first place in the 2009 Document Image Binarization Contest (DIBCO 2009) organised by National Center for Scientific Research “Demokritos”, in Athens, Greece.

Singapore, 26 August 2009 – Ever wished for your Optical Character Recognition (OCR) software to capture a document you had just scanned more accurately? Tired of having to proofread what you had just scanned to make sure that the text matched your hard copy? Well, your wish may have just come true sooner than you think!
Dr Lu Shijian from A*STAR Institute for Infocomm Research (I2R) and Professor Tan Chew Lim from National University of Singapore’s School of Computing have beaten thirty-five other competitors to clinch the first place in the 2009 Document Image Binarization Contest (DIBCO 2009) organised by National Center for Scientific Research “Demokritos”, in Athens, Greece. DIBCO 2009 was held in conjunction with the 10th International Conference on Document Analysis and Recognition (ICDAR 2009) in Barcelona, Spain, in July 2009, an important international forum for the document analysis research community.
The pair’s submission demonstrated clear strengths among thirty-five research groups from all over the world including reputable research organisations from Europe, US, Australia and Asia, that had participated in the competition with forty-three different algorithms (several participants submitted more than one algorithm). Eventually, the best detection performance was achieved by Algorithm 26 that was submitted by Dr Lu S. of I²R and Professor Tan C.L. of NUS School of Computing from Singapore.
The general objective of DIBCO 2009 is to identify current advances in document image binarization using established evaluation performance measures. Document image binarization, a technology often used in OCR software used to drive scanning software is an important element in making document image analysis and recognition possible.
The DIBCO testing dataset consists of five machine printed and five handwritten images resulting in a total of 10 images for which the associated human-segmented (text) images were built for the evaluation. These are document images where the text is manually segmented by human being (instead of computers). The human-segmented images act as the benchmark to evaluate the accuracy of the designed computer algorithms.
The selection of the images in the dataset was made so that it contains representative degradations as in the real world that appears frequently (e.g. variable background intensity, shadows, smear, smudge, low contrast, bleed-through and show-through).
Professor Lye Kin Mun, Deputy Executive Director (Research) said, “Congratulations are in order for Lu Shijian and the team from our Computer Vision & Image Understanding research department for this achievement. We have continued to demonstrate that our in-house developed technologies are able to compete with best of the best in the global arena on real world challenges. I view this achievement as well deserved recognition from fellow researchers over the world for the team at I2R.”
Professor Ooi Beng Chin, Dean of NUS School of Computing (SoC) said, "SoC is glad to share the sense of achievement chalked up by our colleague, Professor Tan Chew Lim at DIBCO. The win at the international event is testament to the cutting-edge research that Prof Tan has been performing in the area of image and text recognition. Our School celebrates with Prof Tan his pursuit of excellence."

Additional information on this year’s DIBCO is available at http://www.iit.demokritos.gr/~bgat/DIBCO2009/bench.... More on ICDAR may be found at: http://www.cvc.uab.es/icdar2009/
- End –