Recognition of Hanacaraka Characters in Old Manuscripts Using Feed-Forward Networks and Elman Recurrent Networks

The Javanese language has a unique set of letters called Hanacaraka characters, which is different compared to the Latin alphabet. Since modern Javanese ethnics of Indonesia don’t use it anymore for formal conversation and education, this language, especially its Hanacaraka characters, begins to extinct. For the preservation purpose of old manuscripts in Hanacaraka characters, we create a system that can recognise Javanese characters automatically from an old manuscript or writing. For this system, we investigated and employed several methods of image processing, features extractions and machine learning for character recogniser. In this paper, we present the result of our investigation of traditional feed-forward neural networks and Elman recurrent networks and comparing their accuracies to obtain the best recogniser. We also compare the results with the accuracies of the probabilistic neural network and induction tree from our previous experiments. From the comparison, we found that Elman recurrent network outperforms the performance of other algorithms, with accuracy more than 97% for data training and 85% for data testing.