Researchers in the US have discovered a new feedback mechanism that allows the human eye to be sensitive enough to see small details in a scene while also being able to detect the large contrast between bright and dark objects. The process involves boosting the output of certain light receptors in the retina while damping down others – and could help with the design of digital vision systems.

The retina of the human eye senses light using about 100 million photoreceptors and a host of neuronal tricks to convert light signals into a useful picture of the world. One of the most basic processes to occur in the eye is contrast enhancement, which allows the eye to resolve bright and dark objects. This is done directly in the retina by a negative feedback mechanism between neighbouring light-sensitive cells.

The first line of signal processing occurs in the star-shaped horizontal cells just below the surface of the retina. Each of these cells receives input from about 100 photoreceptors (rods and cones) on the retina. A photoreceptor responds to darkness with a burst of a neurotransmitter called glutamate. This chemical causes the horizontal cell to depolarize, and the resulting voltage change shuts off the signalling channels of other nearby photoreceptors, preventing them from emitting more glutamate. This has the result of isolating and sharpening the original signal within a circle of quiet – allowing the eye to see a dark object on a bright background and vice versa.

This process is called "lateral inhibition" and helps the eye to detect the edges of objects. The downside of this mechanism, however, is that it also reduces the maximum strength of the optical signal. This should theoretically cause a loss of dynamic range and limit the eye's ability to pick out faint details – something that doesn't happen in a real eye.

Positive feedback at work

Now Richard Kramer, Skyler Jackman at the University California at Berkeley and colleagues at the University of Nebraska and University of Massachusetts have discovered that the eye is able to pick out fine details using an unexpected positive feedback process. They found that cones exposed directly to glutamate, far from reducing their neurotransmitter production, increased their own production of glutamate four-fold.

Upon further investigation, the team discovered that while the glutamate-mediated change in horizontal cell voltage did have a negative feedback effect, glutamate exposure also caused an additional, more subtle change in the horizontal cells, increasing the number of calcium ions in nearby regions. It's thought that this triggers an increase of calcium ions within the cones themselves, boosting the production of glutamate very locally: in just the initially firing photoreceptor and perhaps a few of its immediate neighbours. "This recoups the signal strength lost to negative feedback, while preserving edge detection and contrast enhancement," explained Jackman.

The discovery of a secondary signalling system is particularly surprising because the retina has received a lot of attention from researchers. "The positive feedback circuit is very susceptible to damage, and disappears when the retina is studied using more traditional preparations, involving slices of the retina," explains Jackman. "This may explain why positive feedback has not previously been observed in such a well studied circuit."

Feedback circuits and other processing at a cellular level have several advantages for the central nervous system. First, it saves time because picking out important features helps the brain to interpret what it is seeing quickly, and make appropriate decisions. Second, there is an issue of information flow: 100 million photoreceptors are connected to a mere million nerve channels. This requires a slimming-down of data without losing anything important.

Emulating the eye

It is these virtues of the eye that researchers in digital vision want to copy. Neuroscience and visual prosthetics experts Stephen Hicks at the University of Oxford said, "This finding could well have a positive and relatively immediate effect in computer vision for intelligent systems." He added, "In computer vision we perform a process similar to the eye's lateral inhibition to identify and enhance the edges in a scene before putting them in context." He believes that this work will give researchers "new ideas for implementing a fast approximation of the boundaries between objects in a video feed, which would improve everything from robot–human interactions to video surveillance".

The work is described in PLoS Biology.