University of Notre Dame researchers are developing a neural network to read ancient handwriting based on human perception to enhance deep learning transcription.
Notre Dame's Walter Scheirer says the documents are written in archaic languages and long-unused styles dating back centuries. The project aims "to automate transcription in a way that mimics the perception of the page through the eyes of the expert reader and provides a quick, searchable reading of the text," he says.
Scheirer's team combined traditional machine learning techniques with visual psychophysics, the measurement of links between physical stimuli and mental phenomena. They studied digitized ninth-century Latin manuscripts written by scribes in Switzerland's Cloister of St. Gall, with readers inputting manual transcriptions into a software interface while their reaction times were measured. "We then inform the network of common difficulties in the perception of these characters and can make corrections based on those measurements," Scheirer says.
The team describes its work in "Measuring Human Perception to Improve Handwritten Document Transcription," published in IEEE Transactions on Pattern Analysis and Machine Intelligence.
From University of Notre Dame
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA