Algorithm cracks Google’s Captcha system 67% of the time

It may seem like an unnecessary hurdle while you’re online shopping, but Captcha tests are a key way that websites ensure online security.

The tests, which ask you to identify words or pictures, are designed to tell the difference between man and machine.

But a shocking new study has shown that computers can now successfully crack the test.

This suggests that Captcha tests no longer offer the protection they used to.

 

Captcha tests, which ask you to identify words or pictures, are designed to tell the difference between man and machine. But a shocking new study has shown that computers can now successfully crack the test

CRACKING CAPTCHA

The researchers took inspiration from the brain to develop the algorithm, focusing on ‘contour continuity’ – the way the human brain can distinguish the edges of an object, even if that object is partially blocked by another.

The algorithm they created, called the ‘recursive cortical network’ contains just a few processing layers.

Firstly, pixels in an image are analysed to determine if they form part of an object’s edge.

This data is processed by the rest of the network, and as the data moves through the layers, the image’s contours merge.

The final layers produces the object in question, such as the letter A.

During trials, the algorithm was able to beat Google’s Captcha 67 per cent of the time, PayPal’s 57.1 per cent of the time, and Yahoo’s 57.4 per cent of the time.

In comparison, a Captcha is currently considered useless if it can be cracked one per cent of the time.

Researchers from Vicarious, a technology firm based in San Francisco have developed an algorithm that can be trained to solve complex Captchas, including those used by PayPal and Google.

Speaking to ABC News, Dr Miguel Lazaro Gredilla, one of the researchers who worked on the study, said: ‘We just have to accept that, as computer vision systems improve, traditional text-based Captcha systems no longer offer the protection they used to.’

The researchers took inspiration from the brain to develop the algorithm, focusing on ‘contour continuity’ – the way the human brain can distinguish the edges of an object, even if that object is partially blocked by another.

The algorithm they created, called the ‘recursive cortical network’ contains just a few processing layers.

Firstly, pixels in an image are analysed to determine if they form part of an object’s edge.

This data is processed by the rest of the network, and as the data moves through the layers, the image’s contours merge.

The final layers produces the object in question, such as the letter A.

During trials, the algorithm was able to beat Google’s Captcha 67 per cent of the time, PayPal’s 57.1 per cent of the time, and Yahoo’s 57.4 per cent of the time.

And when the researchers optimised the algorithm for a specific style, it could achieve up to 90 per cent accuracy.

The algorithm contains just a few processing layers. Firstly, pixels in an image are analysed to determine if they form part of an object's edge. This data is processed by the rest of the network, and as the data moves through the layers, the image's contours merge. The final layers produces the object in question, such as the letter A

The algorithm contains just a few processing layers. Firstly, pixels in an image are analysed to determine if they form part of an object’s edge. This data is processed by the rest of the network, and as the data moves through the layers, the image’s contours merge. The final layers produces the object in question, such as the letter A

WHAT ARE CAPTCHA TESTS? 

Captcha (Completely Automated Public Turing test to tell Computers and Humans Apart) has become the standard term for simple human-or-robot tests.

Captchas work by asking you to type a phrase that a robot would be hard-pressed to read.

Commonly, these Captcha phrases are .gif pictures of scrambled words, but can also be .mp3 voice recordings.

These pictures and recordings are very hard for conventional software programs to understand, and robots are usually unable to type the phrase in response to the picture or recording.

In comparison, a Captcha is currently considered useless if it can be cracked one per cent of the time.

Thankfully, other measures are in place in most systems to ensure that Captcha is tricky to crack.

In March, Google ditched its Captcha system in favour of an ‘invisible’ system, that relies on distinct cues, such as mouse movements, to work out if you’re human.

Beyond showing weaknesses in website’s security system, the algorithm could have a range of uses.

Dr Sarah Erfani, an expert in machine learning at the University of Melbourne said: ‘It could be used in medical imaging, where you only have a few samples [of a disease or disorder.

‘From a computer science perspective, it’s wonderful.’

Thankfully, other measures are in place in most systems to ensure that Captcha is tricky to crack. In March, Google ditched its Captcha  system in favour of an ¿invisible' system, that relies on distinct cues, such as mouse movements, to work out if you're human

Thankfully, other measures are in place in most systems to ensure that Captcha is tricky to crack. In March, Google ditched its Captcha  system in favour of an ‘invisible’ system, that relies on distinct cues, such as mouse movements, to work out if you’re human

And the researchers behind the algorithm now plan to apply it to robotic vision.

Dr Gredilla added: ‘In robotics, we need to manipulate objects that are subject to continuous change, in environments that might not be completely controlled.

‘We hope that [the recursive cortical network’s] ability to learn from very few examples will…enable robots that learn and adapt faster to changing tasks and environments.’

 

Read more at DailyMail.co.uk