- Australia's largest state hits record-high COVID-19 cases despite weeks of strict lockdown. 'It's a tinderbox ready to explode,' one official said.
- eToro says crypto made up 73% of trading commissions in the last quarter, as retail customers dived in
- A flight attendant says she was too exhausted to report a passenger who shoved her when a flight was overbooked: report
- US jobless claims climb for first time in 5 weeks, to 353,000
HISTORY IS FULL of manipulated images. Adolf Hitler, Joseph Stalin, and Mao Zedong all were known to cut former allies out of historical photos after a disagreement.
Even Winston Churchill wasn’t above tampering with photographs when it suited him.
Back then, you needed special technical equipment and time-consuming darkroom techniques to manipulate an image.
Since photos went digital, however, it has become a lot easier to modify them, and much harder to tell when an image has been tampered with.
But a Czech company has created a product that can do just that, and banks and insurance companies are using it to detect fraud.
Back in 2004, a group of scientists and digital experts led by Babak Mahdian came together in Prague with one goal: to build a toolkit that could instantly detect whether an image had been tampered with.
They had heard fake IDs were being submitted to banks so people could create fraudulent accounts online, and manipulated images were being used to trick insurance firms.
The team created Verifeyed, which uses machine learning to figure out whether photos have been through editing software and can establish which camera or phone was used to take them.
Traditional digital cameras have several components: an optical system, then a photo sensor, and finally a storage system. Each component has a unique “fingerprint.”
This “fingerprint” can be unique to a camera or camera model, and it can be corrupted when an image has been tampered with.
One of the ways Verifeyed spots fake images is by finding out how many times an image has been compressed into a JPEG file format for storage.
If an image has been tampered with, it is decompressed, loaded onto photo-editing software, manipulated, and recompressed.
JPEG is a “lossy” file format, which means that every time you compress a JPEG image, some information is lost to make a smaller file.
When a JPEG image is compressed, it is split into adjacent blocks of pixels. Those blocks are compressed separately but still have to relate to one another in the same way they would in the original image.
If someone has made changes to parts of the image, the changes will not relate to one another in the same way. The Verifeyed algorithm is able to spot these differences.
Once it was created, Verifeyed took on contracts with a bank, a media agency, and two insurance firms.
After six months reviewing over 1.5 million images, about 2% of the images Verifeyed flagged were false positives — real photos incorrectly flagged as fake.
That isn’t a large percentage. But if an insurance company received 4 million digital images annually, that would mean 80,000 false alarms. A lot of customers could be falsely accused of fraud.
And the Verifeyed product took one or two seconds to process each photo. A company can have thousands of photos to get through daily, so the process took too long.
The team went back to the drawing board. Verifeyed’s clients wanted a system that wasn’t too slow or too sensitive.
After some improvements, only 0.01% of the digital images Verifeyed examines are false positives, and it gets through images in less than a second.
Verifeyed, which now has nine people on its team, did not disclose its revenues to us. Mahdian also says the company has not raised any venture capital funding.
Instead, Mahdian says, Verifeyed has relied on a few business angels, like Jan Stok, the CEO of insurance firm MAI Insurance CEE. The team also uses the resources of its parent company, a computer vision development firm called ImageMetry.
Here are some examples of famous fakes:
A photo of Sarah Palin was went viral shortly after Palin was announced as the vice-presidential nominee for the Republican ticket, depicting her in a patriotic bikini holding a rifle. It was actually a composite of Palin’s head and somebody else’s body. CNN believed it was real for a time.
The Orthodox Israeli newspaper Yated Neeman digitally removed two female Cabinet members from a photo of Prime Minister Benjamin Netanyahu (centre left) and President Shimon Peres (centre right), replacing the women with male Cabinet members.
A Russian newspaper distributed by a pro-Kremlin group printed a manipulated image supposedly showing blogger/activist Aleksei Navalny standing beside Boris A. Berezovsky, left, an exiled financier sought by Russian police.
The Saudi-owned English news website Al-Arabiya published a photo in which fighter jets were digitally inserted above Libyan rebel fighters near a checkpoint on the outskirts of Ras Lanuf in the north of the country.
READ: An Irish woman’s ‘Hailo for cleaners’ has just been bought for €32 million >
READ: Here’s what your kitchen could look like in 10 years >
have your say