The minimum amount of information that the brain needs to recognise an object has been discovered.
Engineers say this could help them build better security cameras which balance the need to identify suspects every time while keeping the resolution as low as possible to cut costs.
鈥淲hen someone raids a bank, the image from the CCTV camera is often so poor you can鈥檛 tell who did it, so it鈥檚 pointless having it at all,鈥� says Alan Lettington of the University of Reading. 鈥淲hat we wanted to know was how good does the picture have to be for you to recognise it.鈥�
Advertisement
To find out, Lettington and his colleague Alison Fairhurst recruited 58 volunteers and showed each of them 500 different images. Each image was of a simple shape such as a triangle or a square, but some were all blurred to varying degrees. Lettington found that when the images were blurred beyond a certain limit, volunteers were unable to distinguish one shape from another.
To determine what that limit was, Lettington used a computer to subtract the images shown to the volunteers from the images they were assumed hold in their memory. This gave an image of the difference between the two. He found the brightness of this difference image was the crucial factor. If it was less than a tenth of the original, the brain was unable to identify it.
鈥淚f you were comparing two faces and the noses were in a different place, when you take one face from the other, you鈥檙e left with the details you鈥檙e trying to discriminate and they have a certain brightness,鈥� says Lettington. 鈥淯nless it鈥檚 more than one-tenth as bright as the original image, you won鈥檛 be able to discriminate between the two.鈥�
Lettington says the limit is set by the level of 鈥渘eural noise鈥�, the background electrical activity caused by neurons firing in the brain. This can drown out weaker signals.
Knowing the least amount of information the brain needs to recognise an image could have practical uses for engineers, especially those making security cameras, says Lettington. 鈥淭hey can work out just how sharp an image needs to be if they鈥檙e to be able to distinguish a face with it.鈥�