Document image binarization is an important step in the document image analysis and recognition pipeline. It is imperative to have a benchmarking dataset along with an objective evaluation methodology to capture the efficiency of current document image binarization methodologies. Following the success of DIBCO and H-DIBCO series (DIBCO 2009 organised in conjunction with ICDAR'09, H-DIBCO 2010 organized in conjunction with ICFHR 2010, DIBCO 2011 organised in conjunction with ICDAR'11, H-DIBCO 2012 organized in conjunction with ICFHR 2012, DIBCO 2013 organized in conjunction with ICDAR 2013 and H-DIBCO 2014 organized in conjunction with ICFHR 2014, H-DIBCO 2016 organized in conjunction with ICFHR16), the follow-up of this contest aims to be organised in conjunction with ICDAR 2017.
In DIBCO 2017 the general objective is to record recent advances in machine-printed and handwritten document image binarization using established evaluation performance measures. The benchmarking dataset that will be used in the contest will augment the existing dataset of the DIBCO series containing machine-printed and handwritten document images that are representative of the potential problems which are challenging in the binarization process.
All researchers in the field of Document Image Binarization are invited to participate in DIBCO 2017. The description of the methods and the evaluation scores will be presented during a dedicated to Competitions ICDAR 2017 session. A report on the competition will be published in the ICDAR 2017 conference proceedings. At the end of the competition, the testing dataset along with the required program to run the evaluation measures will become publicly available similar to the policy followed in the previous competitions of the DIBCO and H-DIBCO series.
NOTE : You may participate in this contest even if you do not plan to attend the ICDAR 2017 conference.
1Visual Computing Group,
Department of Electrical and Computer Engineering,
Democritus University of Thrace.
2Institute of Informatics and Telecommunications,
All registered participants are required to submit one executable:
Binarize arg1 arg2
arg1: Input gray scale or color image
arg2: Resulting binary image
The evaluation methodology will be based upon the latests DIBCO contest, i.e. DIBCO 2013, H-DIBCO 2014,H-DIBCO 2016.
The results will become publicly available at the end of the ICDAR 2017 conference.
with subject: DIBCO 2017 in the registration e-mail, please add affiliation and contact details.