A project providing digital recognition of handwritten documents
using machine learning techniques, delivered via the eScript app
with resources provided via eScriptorium.
A Digital Palaeography Framework
eScriptorium is a project whose aim is to provide a framework within which to organise and transcribe handwritten documents (especially old ones) so that the results of the transcription are available for scholarly and digital use, including searching. It uses the eScript app to deliver this framework.
eScript is an app, written in Java, available on Mac, Windows and Linux, that provides a framework to digitally recognise handwriting, especially as used in old documents.
Its use can be extended beyond old documents to recognise any hand- or type-written document.
It will be appreciated that interpreting old handwriting can be a challenge even for the human brain and so eScript should be seen as an aid to the process of transcription.
It allows you to isolate characters, interpret those characters, automatically interpret similar characters, and add your own interpretation etc in order to achieve a complete transcription of the document. The process of training a neural network and the whole basis of the machine learning environment ensures that where the same handwriting is used a higher degree of automatic recognition should be achieved.
eScript relies on machine learning techniques to provide the recognition mechanism. A neural network (NN) model is initially trained in the handwritten or other characters that are to be recognised and then further characters are supplied to the model with the intention of determining what they represent. The success of the transcription therefore relies on the ability of the NN model to correctly interpret the characters that are supplied to it. This ability is affected both by the quality of the initial training and the inherent capability of the model to correctly apply the training parameters to the supplied characters.
Collaboration is sought to help bring this project to a successful conclusion. Some more work is needed in order to achieve a satisfactory recognition level. At present only the encog machine learning environment is available for use to construct a NN model. We would like to add Tensorflow and use Amazon Web Services as further options.
Contact us to start the dialogue