Images from microscopes or cameras are never perfectly sharp. For 150 years, scientists have known that there is a limit to how clear an image can be, no matter how well the device is made. This limit exists because of the nature of light and how information is transmitted. When trying to measure the exact position of something tiny, like a particle, some blurring always happens. This is not because of bad technology but due to the basic rules of physics.
Researchers from TU Wien in Vienna, the University of Glasgow, and the University of Grenoble wanted to find out the absolute limit of precision possible when using light to measure things. They also wanted to see how close they could get to this limit. They succeeded in calculating the best possible precision and created artificial intelligence (AI) algorithms to help achieve it. These algorithms used neural networks to improve precision in imaging, such as in medical scans.
Neural networks learn from chaotic light patterns
The researchers aimed a laser beam at a small reflective object behind a cloudy liquid. The liquid scattered the light, creating distorted patterns that looked random to the human eye. These patterns made it hard to tell where the object was. The experiment changed how cloudy the liquid was to see how it affected the results. They fed many of these distorted images, each with a known object position, into a neural network. Over time, the network learned to match the patterns to the correct positions. After training, it could accurately find the object’s position even with new, unknown patterns.
The precision they achieved was very close to the theoretical maximum, which they calculated using a measure called Fisher information. Fisher information shows how much useful data an optical signal has about something, like an object’s position. If this measure is low, precise measurement becomes impossible, no matter the method. The AI algorithm’s results were almost as good as the physical limit allows, showing it is nearly perfect. This discovery could improve optical methods in fields like medicine, materials research, and quantum technology. The researchers plan to work with experts in applied physics and medicine to use these AI methods in real-world systems.
This study is published in Nature Photonics.