Scientists develop an algorithm to scan an individual's handwriting and replicate it precisely; no matter how illegible the scrawl. The software could be used to give typed text a more personal touch, and could even help spot forged handwriting. Matthew Stock reports.
STORY: Handwriting is a dying art thanks to the dominance of the QWERTY keyboard. But a new computer program could revive the handwritten word for the digital age. By analysing a sample of a person's handwriting, the algorithm identifies the unique shape and structure of each handwritten letter. It then replicates the typed text as if the author had handwritten it themselves. (SOUNDBITE) (English) TOM HAINES, RESEARCH ASSOCIATE AT UNIVERSITY COLLEGE LONDON (UCL), SAYING: "Essentially it takes individual glyphs; that's the individual instances of each letter you've written. And it stitches them together and it's optimising various cost functions to make sure it looks real. And it's all about getting the spacing right, getting the line to look like a consistent line that was drawn with a real pen. And then stitching it all together so that all the little details that would give away that it's fake are just not there. It's a bit like a magic trick in that respect." The machine learning algorithm adds randomness. So, commonly used words - such as 'and' or 'the' - don't look exactly the same when the system generates them. To demonstrate, the team replicated a famous letter supposedly penned by U.S. President Abraham Lincoln. Such a program could, in theory, make it easier to forge handwriting. But a key goal of the project was helping to spot fakes. The bogus Hitler Diaries caused a sensation in the early 1980s, but wouldn't fool their program, they say. (SOUNDBITE) (English) TOM HAINES, RESEARCH ASSOCIATE AT UNIVERSITY COLLEGE LONDON (UCL), SAYING: "So this project was a stepping stone towards being able to detect questionable, questioned documents. Documents where we don't know who wrote them. So it can be used to extract statistics from a sample where we do know who wrote it and use that to determine if this person actually wrote this document or not. An example of this kind of scenario would be the Hitler diaries which were, back in the 1980s, shown to be fake but at the time it took them a lot of effort and work to determine this." The algorithm has been released as open source software for academic use. But the team also see its commercial appeal. They envision an app-based tool that would let people write personalised letters and greetings cards without ever having to put pen to paper.