Advancing Ancient Script Preservation with a Digital Pen System for Tracing and Recognition

Digitizing ancient scripts is generally achieved using image-based techniques, such as Optical Character Recognition (OCR), or hardware devices, such as stylus-based smart pens. This paper presents a GUI application developed with Python, which allows users to trace illegible ancient scripts on a canvas interface that improves legibility and accuracy of recognition. For character recognition in ancient scripts we implemented a pen interface using a deep learning based Python GUI. We built a pen interface with Python in Tkinter to collect samples of the scripts while the user traced them. The use case was the user was tracing and digitizing ancient scripts to build datasets. After the dataset was collected we used a number of preprocessing techniques such as binarization and noise removal. Then we used a number of different techniques for segmentation (horizontal, vertical, and contour features) to separate the individual characters where we trained the deep learning models. After preprocessing, reshaping the data we trained deep learning neural network models. The best-performing model, of all that were trained, achieved an overall validation accuracy of 94.00%. We also examined real-time examples in application of the trained model, which yielded promising results. The previous accomplishments will provide a strong spring-board for the future, including the expansion of the dataset to incorporate additional characters and additional subject-matter.