top of page
  • Voltaire Staff

Now an AI tool that mimics your handwriting

An image of an ancient scroll

AI researchers have created a tool that could imitate your handwriting style, improving over previous two models that failed in creating meaning in spacing between words, and ligatures, the connecting lines between characters.


Scientists at Mohamed Bin Zayed University of Artificial Intelligence in Abu Dhabi, have created an AI system that learns a person's handwriting style, combining old and new technologies.


Recently patented by the United States Patent and Trademark Office, the tool could assist individuals with injuries preventing them from writing by hand, researchers claimed.


Additionally, it has the potential to generate a significant amount of data, enhancing machine learning models' capability to understand handwritten text more effectively.


Hisham Cholakkal, an assistant professor at MBZUAI and one of the inventors of the technology said, "We wanted to know if you gave a model a few samples of someone’s handwriting if the model could learn about the style of that person and then write anything in the handwriting style of that person."


The researchers said that previous methods of copying a person's handwriting style used a machine learning technique, called a generative adversarial network (GAN).

GANs, could capture the overall style, like how a person slants letters or the width of their strokes, but struggled with creating individual characters and connecting lines (ligatures). Instead of GANs, the researchers used vision transformers, a type of neural network for computer vision.


Unlike GANs, vision transformers can handle long-range dependencies, they claimed, to understanding meaningful relationships even between physically distant parts of an image.


In their study, the researchers compared their method for creating handwritten text images, called HWT, with two other technologies, GAN and Davis et al, and showed the generated text to 100 people and asked which one they liked best. The participants preferred HWT 81 per cent of the time.


Their model doesn't need a lot of data for training; just a few paragraphs of original handwriting are sufficient, researchers said.


One of the inventors, Rao Muhammad Anwer, said the model could be used to decipher doctors' difficult handwriting and even create personalised ads. The tool, not yet available to the public, currently works in English and some French but faces challenges with Arabic.


The inventors are cautious about potential misuse for forging handwriting, as they recognise that handwriting reflects a person's identity.

Anwer said, "Handwriting represents a person's identity, so we are thinking carefully about this before deploying it."




bottom of page