August 2008


This directory includes the ENST-TELECOM ParisTech database of ancient printed characters.

These characters were extracted from an ancient printed book (XVII century) written in French.

The Training set includes about 1,500 characters from 3 clean pages.
The Test_standard set includes 1,106 characters from 2 clean pages 
The Test_degraded set includes 1,191 characters from 2 degraded pages. 
This last set includes many fainted characters. 


Please reference this database with following citation:


L. Likforman-Sulem and  Marc Sigelle, Recognition of degraded characters using dynamic Bayesian networks, Pattern Recognition, Vol. 41, 3092-3103, 2008.



Please note that we provide here the full set of lowercase characters but
the experiments described in [Likforman-Sulem and Sigelle, 2008] use only 
16 character classes (the most populated ones).