Babak Mahdian is an award-winning scientist and businessman who heads ImageMetry – a company which developed cutting-edge software called Verifeyed for use by insurers, law-enforcement agencies, and other clients to verify the authenticity of digital images and uncover digital fraud. In an age when any image on the internet can be manipulated within minutes or seconds, knowledge of the source – its digital fingerprint – as well as the sure-fire ability to uncover tampering is more important than ever.
“Well I love science and I love image processing and computer-image tasks but I also have a passion for business. So I always wanted to try to find a way to be active in the both: business but also being able to solve complex scientific tasks. The only way to do this – the only effective way to achieve this dream – was to found a company focussing on image processing applied to business and industry.”
Starting a company is always a difficult process... what were the company’s beginnings like?
“I had every confidence in a good start: we had great know-how and a very unique team in terms of understanding applied mathematics, image processing, artificial intelligence, and so on. So I expected an easy start taking our products to the market, to acquire new clients and so on. But I have to admit that the first two years didn’t go the way I had expected. We weren’t really successful at all.
“The reason is that being a great scientist does not necessarily make you a good businessman. What makes you good in business is listening to clients, hearing what they want, understanding what they need. In the beginning, we had it the other way around: we offered certain services to customers but what we gradually learned to do was to listen, to their problems, and then to provide solutions. Since we changed that over our five-year existence, we got bigger and bigger and far more successful.”
You have enjoyed quite a measure of success with the project Verifeyed: what is the focus with that product?
“Permit me a brief history about the digital camera: the first was developed by Eastman Kodak back in 1975 and originally had uses only in science and the military. Several years later it also began to play a role in medicine. In the 1990s, we saw the first attempts to bring the digital camera to the consumer market and replace the traditional film camera. The beginning was quite had but I would say that from the year 2000 onwards digital replaced film. The focus was to bring a better digital camera, more user-friendly, with a lot of functionalities and a very nice user experience. But what the industry forgot about was the integrity of digital images because nobody expected digital images to become, so quickly, so widespread in every industry.
“It is so easy to take a photo, to send it somewhere, to publish it. The problem is that digital editors like Photoshop and others also make it very easy to change photos, and to change the information provided. For example, today you can very easily, using a few clicks, fake a car accident, or personal I.D. or a driver’s licence. And that’s a problem. Often insurance companies will pay out a claim based on photo evidence but if you can’t ‘trust’ a digital photo... That was the reason why we started Verifeyed. To provide a solution to an existing problem and bring it to the market.”
In a nutshell, we are talking about developing software which would quickly – and reliably – determine the authenticity of individual pictures...
“Exactly. Our aim was to produce software which, with one click, would tell the user if the photo had been tampered with or not.”
How does it actually work?
“Traditional digital cameras consist of several components: you have an optical system, then a photo sensor, and finally a storage system. Light is projected to the photo sensor, the sensor captures the intensity of the lights in individual segments called pixels and in the end they are stored in imaging formats like Jpeg. What is important here is that we, and not just us but also other scientific labs, discovered that EVERY component brings very unique artefacts, maybe we can call them fingerprints, to each photo.
“Sometimes these are unique per camera, and sometimes they are unique to the model of the camera. And when you digitally alter a picture, the information becomes corrupted. Using this knowledge, we were able to build up an algorithm which extracts the fingerprints from photos and determines whether they are built up by the camera or by a photo editor like Photoshop. Generally-speaking, that is how we find out if a photo is original or has been tampered with.”
Historically, it’s nothing new that photographs would be tampered with. If you look back to every dictator of the twentieth century, many official photographs were changed, sometimes many times over. Fellow party members disappearing from photos next to Stalin. But given the speed at which images are spread today, the need was much greater.
“You are absolutely right; we can see a lot of image tampering created by the supporters of Stalin or Hitler or various Chinese dictators. But whereas in the past the main goal was political, today we see an increasing financial motivation behind image forgery, with forgeries created by criminals easily abusing, for example, Photoshop.”
Are mainstream programmes like Photoshop the ones most often used for this kind of fraud, or are there more sophisticated kinds of systems out there?
“That’s a good question. When we detect tampered photos in various industries, we see there are typically two main kinds of forgeries. The first type, I would say, are the ‘naive’ people not having a great deal of knowledge about how to forge photos, and use Photoshop and simply change something in the photo and submit it to an insurance company.”
Putting a dent in the car or a hole or a broken windshield, that kind of thing?
“Exactly. And also today we see a lot of fraud in online loans.”
So those are the amateur users, but you were also saying there were more serious cases.
“Yes, especially in the financial and insurance industries. Today there is organised crime focusing on forgery creation and digital fraud. These kinds of guys do not use Photoshop, they are quite smart. We see a lot of such fraud coming from Eastern Europe, they have perfect knowledge in photo tampering and use more advanced digital image editing software.”
So what is the state of image forensics now? Are you keeping pace with the most sophisticated frauds?
“First of all I have to say that it is a never ending game between forgery creators and forgery detectors. But today I can claim that we are able to detect most of the digital image forgeries. For example, in the insurance industry, one out of around 800 of insurance claims, especially in the outer insurance industry, contains a digitally manipulated photo. How the market and the situation will look like in 2 years time, I’m not sure, but I would say that we are doing our best to minimise this type of crime.”
The area of application for software like Verifeyed is varied. What are some of the most serious areas where it can be used?
“Regarding the detection of photo tampering, it would be insurance, financial and forensic investigation. But when we talk with intelligence or security organisations to understand what they need, we have been told that one of the biggest problems they have are for example terrorist organisations, who sometimes publish a video or a photo for blackmail purposes.
Or often there are times we see photos or videos of child sexual abuse on the internet: an unquestionably great problem in our society today. So the need is to understand who made these images. Having this information available can be very helpful in arresting these people and stopping further crimes of this sort. Inspired by these reasons, we at Verifeyed are working on being able to tell you who the author of any published photo on the internet is, even if the photo is published anonymously.
A lot of Facebook’s one billion users publish photos on their webpage, and we see the same on Flickr or Google. So having these photos available, we can create a so called digital fingerprint of these people and their source camera. So when we find a photo on the internet which has been published anonymously, having this information about digital footprints available, we can extract the features we need from the photo and match them to our database of digital fingerprints and actually identify who the author of the photo is. I believe this technology can have a very high value in the fight against child sexual abuse and terrorism.”
“That’s our first problem. Our research and development is focused on this at the moment. We also have at the moment a patent pending algorithm combining parallel computation, artificial intelligence and image processing methods to solve this problem. So today, in less than one second, we can identify the author of a photo in a database of one hundred million digital fingerprints. In real time.”
The episode featured today was first broadcast on October 22, 2013.