The computer program perfectly reproduces your handwriting


It’s a common topic of conversation: the dying art of handwritten letters and more personal forms of communication. The point is, typing communication is very effective. Typing and pressing send is faster than writing a note or letter and then sending it to someone, and texting can often convey important information much faster than a phone call.

It doesn’t mean that the slower versions of things have lost their place in this world, they just need to be reimagined. Scientists at University College London have developed a computer program capable of reproducing a person’s handwriting, called “My Text in Your Handwriting”.

The program examines a person’s handwriting – a sample as small as a paragraph – and then allows the user to enter new text using their own handwriting. The art of handwriting can continue, but with typing and email speed.

“Our software has a lot of valuable applications,” said Dr Tom Haines, professor of computer science at University College London and one of the developers of the program. “Stroke victims, for example, may be able to write letters without worrying about illegibility, or someone sending flowers as a gift might include a handwritten note without even going to the florist. be used in comics where a piece of handwritten text can be translated into different languages ​​without losing the original style of the author.

The program is yet another example of machine learning software. We’ve seen examples of machine learning in a variety of things like self-driving cars, the Nest smart thermostat, robots learning to do household chores, and apps that can identify flowers.

In this case, the algorithm analyzes a person’s glyphs or the way a person writes a specific letter. The software finds what is consistent in the style and spacing of these glyphs, and then reproduces it. Beyond just characters, the software reproduces the texture and color of a person’s pen line, ligatures (marks joining letters) and vertical and horizontal spacing.

Researchers say that unlike fonts that mimic handwriting but are clearly computer generated, the software produces text that actually looks like handwriting. They tested this by asking people to distinguish between envelopes that had been addressed by their software and those that were written by hand and volunteers were deceived by the software 40% of the time.

The team took famous handwriting samples and reproduced the calligraphy of such luminaries as Abraham Lincoln, Frida Kahlo, and Arthur Conan Doyle (whose original handwriting is pictured above, followed by the software version). While some might say it could lead to counterfeits, the researchers say their handwriting analysis software, which synthesizes tiny details in texture and shape, might actually be a useful tool to help identify counterfeits.


Gordon K. Morehouse