A New Text-Independent GMM Writer Identification System Applied to Arabic Handwriting

This paper proposes a system for text-independent writer identification based on Arabic handwriting using only 21 features. Gaussian Mixture Models (GMMs) are used as the core of the system. GMMs provide a powerful representation of the distribution of features extracted using a fixed-length sliding window from the text lines and words of a writer. For each writer a GMM is built and trained using words and text lines images of that writer. At the recognition phase, the system returns log-likelihood scores. The GMM model(s) with the highest score(s) is (are) selected depending if the score is computed in Top-1 or Top-n level. Experiments using only word and text line images from the freely available Arabic Handwritten Text Images Database written by Multiple Writers (AHTID/MW) demonstrate a good performance for the Top-1, Top-2, Top-5 and Top-10 results.

Published in:
Proceedings of 14th International Conference on Frontiers in Handwriting Recognition, 708-713
Presented at:
14th International Conference on Frontiers in Handwriting Recognition (ICFHR 2014), Crete Island, Greece, September 1-4, 2014

 Record created 2014-10-03, last modified 2019-08-12

Rate this document:

Rate this document:
(Not yet reviewed)