GREC-SEG Database

Introduction

The main goal of this database is to provide a framework for the evaluation of different symbol recognition methods dealing with vectorial primitives.
The symbols of this database are a subset of 80 different symbols from the original International Symbol Recognition Contest GREC'2005. The used symbols are the ones which are only composed by straight lines and have no arcs. For each class 60 distorted instances have been generated.

The vectorial representation of each symbol is represented by an attributed graph where the nodes represent points and the edges a segment between two endpoints. Each node from the graph is randomly shifted within a predefined radius r. Three different levels of distortion are generated with values of r equal to 5, 10 and 15, respectively. At each level, 20 different instances of the symbol are generated. Notice that the graph representation aim to maintain the connectivity and the number of segments which compose the graphical symbol.



Database Description

The GREC-SEG database consist of 80 different folders:
  • Each folder corresponds to one of the model symbols.
  • In each folder there are the .bmp and the .vec representations of the symbols.
  • Three different levels of degradation have been used. The lowest level corresponds to the files symbolXXX_YY_05.vec, the intermediate level corresponds to the files symbolXXX_YY_10.vec, the highest level corresponds to the files symbolXXX_YY_15.vec.



Download the GREC-SEG database

All the data is publicly available and can be downloaded. We invite you to use these vectorial images and the models for research in symbol recognition (for scientific, non-commercial purposes only). In this case, we would appreciate acknowledgement of the source. Please make the reference to the following paper
  • Symbol Spotting in Technical Drawings Using Vectorial Signatures.
    M. Rusi˝ol and J. Lladˇs. In Graphics Recognition. Ten Years Review and Future Perspectives, volume 3926 of Lecture Notes on Computer Science, pages 35-46, 2006.
    [ Paper]   [ 10.1007/11767978_4]


Download the complete database.

This page is best viewed with Mozilla Firefox - Last Updated September 2009
ę 2009 Maršal Rusi˝ol