| 

iapr

GREC 2005 : Arc segmentation

Arc segmentation GREC 2005

We are pleased to announce that the contest on arc segmentation will be held at the GREC'05 workshop, Hong Kong, August 25-26, 2005.

This will be the sixth in the series of graphics recognition contests organized by the International Association for Pattern Recognition's technical committee on graphics recognition (IAPR TC10). The first contest, held at the GREC'95 workshop in University Park, PA, focused on dashed line detection. The second contest, held at the GREC'97 workshop in Nancy, France, attempted to evaluate complete (automatic) raster to vector conversion systems. The third contest, held off-line associated with the GREC'99 workshop in Jaipur, India, also aimed to evaluate complete (automatic) raster to vector conversion systems. The fourth (at GREC’01 Kingston, Ontario, Canada) and the fifth (at GREC’03 Barcelona, Spain) focused on arc segmentation.

The contest this time will also on arc segmentation, which is an important and difficult problem in the field of graphics recognition. This contest will test the abilities of participating algorithms/systems to detect arcs from raster images. We plan to develop and use a set of performance metrics based on the published line detection performance evaluation protocol [8] to evaluate and compare the participating algorithms/systems on-line at the workshop site with test data of different quality and complexity. Pre-contest training images and the performance evaluation software are provided before the contest so prospective participants can try their systems and improve them for optimal performance. Test images could be synthesized and/or real scanned images. Although the contest is on arc segmentation, other line types and/or text may also appear in the test images to increase the complexity. We welcome participations of both commercial software with arc segmentation functionality and the academic research prototypes of arc segmentation algorithms. As has been the case with previous graphics recognition contests, we will enable anonymous participation. A participant can choose to remain anonymous either from the outset or in the contest report, which will be published as a paper that evaluates the performance of the various algorithms following an accepted protocol.

Contest home page : http://www.cs.cityu.edu.hk/~liuwy/ArcContest/ArcSegContest.htm

A preliminary test image and its ground truth vector file are also ready for download.

More test images with various levels of noises and their ground truth vector files are also available.
Click here to download the zipped package. The method used to generate these noises is described in a paper.