In this paper, we present the synthesis of the first international symbol recognition contest, held during the fifth IAPR workshop on graphics recognition in Barcelona. We describe the framework of the contest (goals, kind of symbols, criteria of evaluation) and how the contest was organized. We expose in particular the way we have built the symbol database and which methods have been used to introduce noise in the model of the symbols, in order to evaluate the robustness of the recognition methods. The methods of the different participants are summarily described as well as the results obtained with these methods on the test images available during the contest. The evaluation protocol is presented, the results are analyzed and we conclude with some remarks for the next symbol recognition contest. © Springer-Verlag 2004.
|Journal||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Publication status||Published - 1 Dec 2004|