Introduction

This is the first major competition involving hand-written Tamil characters, and the first release of a public ground-truthed Tamil dataset. See these Wikipedia entries for more information on the Tamil Language and Tamil Script.

A particular challenge compared to other OCR datasets (e.g. such as MNist), is the relatively high number of classes (156), and the relatively low number of training samples per class.

• The objective of the competition is to recognize 156 different classes of handwritten Tamil ‘characters’.
• Participants should register for the competition at their earliest convenience.  Registration indicates an intention to enter rather than a firm commitment. 
• Training data and subsequently test data for the problem will be provided to registered participants in the following formats:
  • Online data in UNIPEN 1.0 format
  • Offline data as bi-level TIFF images derived from online data.
• Recognition results on the test data will be accepted as plain text files in a standard format for evaluation (see here).
• We also encourage submission of a brief system description (plain text or PDF), but this is not essential.
• Participants not submitting results by the deadline will be automatically disqualified from the competition.
• The criterion for evaluation is the highest top-choice accuracy (averaged across all classes) at zero reject rate.
• Results of evaluation will be communicated to all participants within one week of submission of results on test data
• The winners (the N best results, N to be decided based on participation) will be informed individually via email.
• The entries will be ranked regardless of whether they used the on-line, off-line, or both types of input data, though this information will be available in the league table.
• Results for all participants, and awards for the winning entries, will be presented at the 10^th Intl. Workshop on Frontiers in Handwriting Recognition (IWFHR-10) to be held at La Baule, France, between Oct 23-26, 2006
• Winning entries will be invited to present posters on their techniques at a special session at IWFHR, corresponding papers will also be published in a special supplement to the proceedings.
• Submission of paper and poster is optional and left to the participant.
• The organizing committee’s decision is final

If you intend to enter then please register.

Simon M. Lucas
IAPR TC5 Chair

Sriganesh Madhvanath
HP Labs India

sml@essex.ac.uk