Autonomous Perceptive Systems
The automatic recognition of historical, handwritten manuscripsts is not a solved problem. Powerful commercial sorting systems do exist in closed applications, such as address reading of scanned handwritten envelopes. However, historical documents present a huge challenge: the noisy images, unknown character shapes, unknown words and unknown linguistic statistics make the use of traditional approaches impossible. We have investigated how to process historical documents and their transcriptions in such a way that a continuously learning, high-performance computer could learn how to search words in such documents. The result is an advanced architecture, MONK, which allows volunteers and interested human users to enter text labels over the internet. Read more.