ICFHR 2016 Competition on Recognition of On-line Handwritten Mathematical Expressions (ICFHR-CROHME-2016)

Research Tasks


2017-07-18 (v. 1)

Contact author

Harold Mouchère

University of Nantes


+33 2-40-68-30-82


This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.


Symbol Recognition: Classification of isolated math symbols

Lot of expression recognition system use an isolated symbol classifier in their complete process. This task compares them. Furthermore, this simple task is a good simple pattern recognition task to compare machine learning solutions.

This task is split into two sub-tasks :

  • in Task-2a only valid symbols are provided as input
  • in Task-2b incorrect symbol segmentations (junk) are mixed with valid symbols.



For  these tasks, each inkML test file contains a single symbol with an associated identifier (‘UI’ tag). Participants submitted a CSV  file  containing  one  line  for  each  test  file.  Each  line provides the symbol identifier followed by a ranked list of the  Top-10  classification  candidates. 

A  CROHMELib  tool (evalSymbolIsole.py)  was  used  to  compute  symbol recognition rates along with the average rank of the correct symbol class (TMP: True Mean Position), where any target class  not  appearing  in  the  Top-10  is  treated  as  rank  11.

For Task 2b where some inputs were ‘junk’ symbols (i.e.incorrectly segmented symbols), the tool also provides the symbol true positive rate (TAR: True Acceptance Rate) and false positive rate (FAR: False Acceptance Rate).


No comments on this dataset yet.

Add your comment

In order to comment on a dataset you need to be logged on
Register Now!


In order to rate this dataset you need to be logged on
Register Now!