Abstract:
Recognizing the content in the ancient inscriptions unlocks many gateways to the
undiscovered historical events since inscriptions were used as a major communication
mechanism in ancient Sri Lanka. Currently these inscriptions are read through naked
human eye with a great effort. This manual process is not only time consuming but
also can generates uncertain outputs sometimes due to the noise that is available in the
inscriptions.
We hypnotize that the noise removal of a textual document can be resolved through
communication among lexical, structure analyst and semantic agents of a multi agent
solution. This is inspired by the real world scenario where noisy outputs can be
resolved by experts through their knowledge in morphology, sentence structure and
semantics of a particular context.
This thesis is an attempt to recognize the Brahmi characters in ancient Sri Lankan
inscriptions. The overall solution comprises of several agents namely: artificial neural
network agent, lexical agent, structure analyst agent and semantic agent. The input for
the proposed system is an ancient Sri Lankan inscription, this particular inscription
image is pre processed using different image processing techniques and segmented
into isolated characters. The artificial neural network agent analyzes the pixel
intensity of the isolated characters, extract the features and recognize the relevant
Brahmi character using the trained neural network. The recognized character string
could contain Brahmi characters which have identified erroneously due to the high
noise availability. The lexical, structure analyst agent and semantic agents plays a
major role to correct the mistakenly identified characters by communicating with each
other. The output of the system consists of relevant Sinhala Unicode characters for the
recognized Brahmi character string.
Experiments were carried out to evaluate the recognition rate of the system by using
12 inscriptions that were found in archaeological sites – Wessagiriya , Handagala
Vihara etc. 84% of inscriptions were completely recognized and among the rest 8% of
inscriptions were partially identified.
