Fake or real signature? FEEC BUT scientists develop software for forensic examination of handwritten text
Radim Burget cooperates with Prague criminalists on forensic analysis of handwriting | Autor: Jan Prokopius
Handwriting recognition is still one of the most important forensic techniques for obtaining evidence. The Institute of Criminalistics therefore approached experts from the FEEC BUT to collaborate and are now developing software for handwriting analysis based on artificial intelligence and computer vision. It should help detect fake signatures or threatening messages. The project started this year and should be finished in 2025.
Similar handwriting recognition software already exists abroad, but because of the Czech diacritics and the specific way of writing, criminals cannot use it.
"Moreover, the font evolves over time and the techniques for newer types of writing, such as cane, have not yet been developed. It would not be feasible to do so using the classical approach, either for time or financial reasons. By combining letter recognition and software automation, we can carry out a very reliable analysis in a few seconds, which would normally take a person a week," explains project leader Radim Burget from the Institute of Telecommunications at the FEEC BUT.
The three-year project is divided into several stages - in the initial one, the researchers are now training a machine learning algorithm to determine as accurately as possible the type of pen used to sign the document. "Using special equipment, we are able to determine the spectrum of a given ink and use artificial intelligence to pair it with an identical ink in the database. At the same time, we are finding out how the properties of a single ink change over time, which is applicable to the analysis of older documents," describes Burget.
The emerging software can then confirm whether multiple documents have been signed with the same ink and by the same person. This is useful, for example, for cases where a person has signed one contract but denies signing others. "The aim is to analyse whether the stiffness on the documents being examined is identical. Moreover, if the person in question used a pen with a unique refill to sign, the importance of such evidence increases. Similarly, if officers secure a pen in the office of the person under investigation, they can link it to the suspect documents. For the court, this can become circumstantial evidence of the perpetrator," explains the project leader.
In the next stage, the researchers will look at writer identification for handwritten handwriting, which will allow the handwriting to be analysed using measurable parameters. "The expression of the scribe will thus be supported by data. Because now if he says he finds two signatures identical and the defence disagrees, the court will exclude the evidence. That's why we're working on developing measurable parameters that will evaluate the type of handwriting, and the AI will then evaluate how the samples are similar or different," Burget says, noting that the software must have enough statistical plausibility for judges to accept it as a credible source of information.
This is not the first collaboration with Prague criminologists for the experts from FEEC BUT - in previous years, for example, they have developed software to recognise blank cheques that were first signed and then overprinted with the amount of money. The aim of the cooperation is to equip the Criminal Investigation Office with a set of software tools and methodological procedures that will make forensic letter examiners more efficient and expand their capabilities.
