Scientists from the Department of Optics at the Faculty of Science, Palacký University Olomouc, have developed a new method that uses artificial intelligence to extract details from blurred images that are practically invisible to conventional optical instruments. The same artificial intelligence model can process images of distant stars as well as individual molecules, without needing to be adapted to a specific telescope or microscope. The results of the research were published in the prestigious journal Nature Communications.
Every telescope or microscope has its physical limits. If two objects are too close to each other, their images merge into a single blurred spot. This complicates the work of astronomers observing distant stars, for example, or biologists studying structures inside cells. Existing methods that try to overcome this limitation require very precise knowledge of the properties of a specific instrument and its careful calibration.
AI learned to see beyond the limits of optics
The Olomouc researchers chose a different approach. They trained the artificial intelligence exclusively on computer-generated simulations covering a wide range of optical conditions. As a result, the model did not learn to work with only one specific instrument, but instead captured the general principles by which images are formed in different types of optical devices.
“Our goal was for the AI to learn not just one specific experiment, but a general way of extracting sharper information from a blurred image. This is exactly why the same model can work with data from very different fields – from astronomy to molecular microscopy,” explained Dominik Vašinka from the Department of Optics at the Faculty of Science.
From laboratory points to stars in Andromeda
The scientists first tested their method in a laboratory experiment where they precisely knew the true positions of the light sources. They then successfully applied it to real data. In images of a region of the Andromeda galaxy, the artificial intelligence was able to distinguish a group of stars that appeared as a single object in the original image. In biological samples, it helped reveal fine cellular structures formed by individual fluorescent molecules.
“The experiment with a controlled true structure of the object was an important step. In real microscopic data, we do not know exactly where the molecules are located, and the situation is similar for astronomical data. We created an intermediate step between an ideal simulation and real samples,” said Filip Juráň from the Department of Optics.
Stars and molecules are connected by the same principle
According to Jaromír Běhal, seemingly unrelated fields are connected by the same physical principle. “The connecting element is not the stars or the molecules themselves, but the physics of imaging point sources. If AI captures this physics in a sufficiently general way, it can help in situations where classical calibration is time-consuming, unstable, or practically impossible,” he said.
The researchers believe that their approach will make it easier to use super-resolution methods in a wide range of scientific and technical fields. Because the model is not dependent on a single specific instrument, it could in the future find applications wherever it is necessary to extract as much information as possible from optical images.
One model for different instruments
“This is not merely another image-processing algorithm. What matters is that the model is not tightly bound to a single instrument. This could significantly simplify the use of super-resolution methods in both research and technical practice,” added Miroslav Ježek from the Department of Optics.
The research builds on the long-term work of the Laboratory of Quantum Optics at the Department of Optics, Faculty of Science, Palacký University Olomouc, in the fields of optical measurement, artificial intelligence, and the processing of weak light signals. The project was supported by the Czech Science Foundation, the Ministry of Education, Youth and Sports of the Czech Republic through the OP JAK ITI Intersectoral Cooperation “Optical Technologies” program, and the Internal Grant Agency of Palacký University Olomouc.
Links:
Article published in Nature Communications: https://doi.org/10.1038/s41467-026-75584-7
Laboratory of Quantum Optics, Department of Optics, Faculty of Science, Palacký University Olomouc: https://opticsolomouc.org/
Illustration author: Monika Tomanová, https://monikatomanova.art/, Instagram @monikatomanova.art