Prince George may still only be a toddler but computer scientists say they can now offer a glimpse of how the third in line to the British throne will look in the coming years. Their new software takes a number of specific facial features and combines it with visual cues from the parents and other relatives to create a detailed portrait of an individuals likely future appearance. The results are far more accurate than existing ageing software, say the scientists from the University of Bradford.
The research, led by Hassan Ugail, the Universitys professor of visual computing, explained that their algorithm can verify which traits the child may have inherited from the parents to build a more reliable forecast of their future face. He said that the software is trained to be more accurate as further layers of data are added. To demonstrate the system, the team chose to predict what Prince George will look like over the coming years, up until the year 2073 when hell celebrate his 60th birthday.
We take specific facial features. Very simple things, like nose length is quite unique for that person, so we look at nose length, the width of the nose, the distance between eyes. So these are facial features that the computer recognises as the person. So we take these - roughly 30 to 40 facial features we take from the face − and we use these facial features; we map it into the machine and then we produce the age, Professor Ugail said.
So what weve done in the case of George, weve taken his picture and then weve actually taken facial features and then aged him. Weve also, in some experiments, what weve done is weve taken the parental information and then also applied the parental information and aged him as well.
The team also created portraits of Prince Georges little sister, Princess Charlotte, ageing her from the age of two up until age 60. Ugail explained how the system assumes a natural age progression; not taking into account the impact of factors like diet and environment. Nevertheless, he believes that the programs facial predictions offer an accuracy of around 80%.
Its very, very difficult to 100% say this is what the person is going to look like because there are other things that come into it. You know, theres environmental issues, there are dieting habits. So all these things can age people very fast. So its very, very difficult to predict it, but what we assume is a natural age progression. So from that point of view our software is roughly 80% accurate, he said, adding that the hair styles on their images were purely for illustration.
The team has been approached about making the software into an app. While Ugail admits there is a fun side to the system, hes keen to highlight its potential in police investigations. Originally, it was developed to identify criminal suspects in crowds, and Ugail believes it could be an important tool to help trace missing children and adults.
We can actually look at a childs pictures, we can incorporate all the pictures available, we can also add information from parents, parental facial features, grandparental facial features if available and also other relatives in the family. So we can take all that information so the accuracy that we can predict would be much, much better compared to the current existing technology, said Ugail.
