We assemble a word model for each training word possibly using several identical copies of the initial character ... In the first two cases, we used machine- printed Arabic characters as templates when manually creating character skeletons.
|Title||:||Proceedings 2005 Symposium on Document Image Understanding Technology|
|Author||:||David Doermann, University of Maryland/UMIACS|
|Publisher||:||UMD - 2005-01-01|