At the GREC 2007 Workshop it was discussed to make the Arc Segmentation contest an ongoing contest, i.e. we will allow submissions of results at any point in time. (Eventually, we might have to switch to new and more test images.) The results of all submissions are shown in the following table. If you have results you would like to include, please send an email including your results to Faisal Shafait. (We prefer results in image format - see below - but VEC format is also acceptable.)

The VEC-files for the commercial systems were supplied with a constant line width of one pixel for all the arcs. To see the influence of this choice, we re-ran the evaluation using a constant line width of 10 pixels for all systems. The following table shows the results.

We have also prepared a discussion page for the results and procedure. What follows is the original page for the GREC 2007 Workshop contest, which also explains the evaluation and links to the data and web services.

IUPR will be organizing the GREC 2007 Arc Segmentation Contest. As in past years, there will be a competition regarding the detection of arcs in digital images held at the Graphics Recognition Workshop GREC (see the homepage of the GREC 2005 contest and its results).

• 2007-03-22 web-site and examples on-line
• 2007-06-27 conversion web-services and training images available
• 2007-08-27 test images available
• 2007-09-03 submission of test results (by email to Daniel Keysers) Extension: the deadline has been extended, please send me an email if you would like to participate.

Here is a tar-gz-file with five scanned images (in pbm format) and the corresponding arc ground truth (in ppm format): train.tar.gz

Here is a zip-file with five scanned images (in tif format, g4-compression) and the corresponding arc ground truth (in tif format, flate compression): train.zip

Here is a tar-gz-file with five scanned images (in pbm format): test.tar.gz

Here is a zip-file with five scanned images (in tif format, g4-compression): test.zip

The web-service for evaluation of results is now available at http://demo.iupr.org/segeval/.

The web-service for conversion of results from VEC-format to the color-based format is now available at http://demo.iupr.org/vec2img/.

Note: the VEC-file may only contain lines that describe arcs or circles (starting with 'A' or 'C', respectively).

The competition will be using a color-based evaluation method for page segmentation. The foreground pixels corresponding to each arc will be assigned a unique color. This results in a color-coded representation of image segmentation. The segmented image thus obtained will be compared to the ground-truth image using bipartite graph matching. Details of this representation and evaluation are described in this paper here. Two example images with their color-coded segmentation representations are shown in the Figure below.

Note that the assignment of colors to arcs is arbitrary, so you can choose any colors for representing different arcs. However, three special color values are recognized:

0xFFFFFF page background

0×000000 un-classified page foreground

0xFFFF00 pixels belonging to more than one arc

If you have any questions, please contact Faisal Shafait.