Sunday, April 28, 2013


The edited-version* of training images for arc segmentation contest is available:


edited-arc-seg-challenge-training-images.zip
edited-ground-truth-arc-seg-challenge-training-images.zip

* Firstly, now training images only contain GREC Arc-Seg Contest 2009 training images. The GREC Arc-Seg Contest 2011 training images have been removed. Secondly, two versions of training images are provided now: images with text and images without text. Both of these types of images are now considered as two sub challenges under Arc segmentation contest.


Sunday, March 10, 2013


Seventh International Line and Arc Segmentation Contest
Tenth IAPR International Workshop on Graphics Recognition (GREC’13)


Contest Theme
"Line and Arc Segmentation of Engineering Drawing and Cadastral Map Images."





Introduction:

The Line and Arc Segmentation Contest 2013 is the seventh in the series of contests and to be held during GREC'13 in Lehigh University, Bethlehem, PA, USA - August 20-21 2013. It is going to be organized by Technical University of Kaiserslautern (Germany), The University of Western Australia, Perth, University of Duhok, Kurdistan Region, Iraq, and University of Science, Malaysia. The most recent contest was held during GREC'11, Chung-Ang University, Seoul, Korea, September 15-16, 2011, and was organized by School of Computer Sciences, University of Science, Malaysia (USM). Previous editions of the contest focused exclusively on arc segmentation from engineering drawings. In the current contest, segmentation for both arc and line will be tested for engineering drawings as well as cadastral maps.

For queries regarding the contest, please contact:

Syed Saqib Bukhari
Technical University of Kaiserslautern, Germany
IUPR - Image Understanding and Pattern Recognition
Email: bukhari@iupr.com



Database:

Engineering drawings (captured from books and record rooms) and cadastral maps (captured from record rooms and gathered from Internet) will be used. The engineering drawing images contain both lines and arcs and the cadastral map images contain only lines. Altogether, the database will contain both non-broken and broken types of lines and arcs, where broken lines and arcs occur due to image degradations. The annotations/text will not be removed from the scanned images, however the annotations/text will not be ground-truthed. Manual ground-truthing will be performed on the binarized images and the ground-truth data will be stored as VEC files. The database will be uniformly split into training and test images.

Performance Evaluation Method:

The current contest will be broken down into two challenges:

  • Challenge-1: line segmentation
  • Challenge-2: arc segmentation.

The standard VRI criterion [1] will be used for performance evaluation of arc segmentation challenge and the pixel-based performance evaluation [2] will be used for line segmentation challenge.

Important Dates:

  • Training images available: 10th April 2013
  • Ground-truth of the training samples available: 20th April 2013
  • Submission of results by contestants: 15th May - 31st May 2013

Organizers:
  • Syed Saqib Bukhari, bukhari@iupr.com, Technical University of Kaiserslautern (Germany)
  • Hasan S. M. Al-Khaffaf, hasan@iupr.com, University of Duhok, Kurdistan Region (Iraq)
  • Abdullah Z. Talib, azht@cs.usm.my, University of Science, Malaysia (Malaysia)
  • Thomas M. Breuel, tmb@iupr.com, Technical University of Kaiserslautern (Germany)


Contact Person: Syed Saqib Bukhari (bukhari@iupr.com)



References:


[1] L. Wenyin and D. Dori, A protocol for performance evaluation of line detection algorithms, Machine Vision and Applications, 9(5/6):240–250, 1997.
[2] F. Shafait, D. Keysers, and T. M. Breuel, Performance evaluation and benchmarking of six-page segmentation algorithms, IEEE Transactions on Pattern Analysis and Machine Intelligence 30 (6) (2008) 941-954.