Directly imaging exoplanets can reveal details about their atmospheres and surfaces, but stars’ overwhelming brightness and imperfect telescope optics make the task difficult.

Researchers at Northwestern University developed FLIP, a physics-based image-processing technique that removes residual starlight from James Webb Space Telescope images and improved sensitivity to faint planets by two to 15 times in tests on existing observations.

Conventional methods subtract light using an image of a separate reference star, but even tiny shifts in telescope alignment and mirror shape can make that reference a poor match for the glare in the target observation.

Instead, FLIP builds a digital twin of the telescope using a physical model and wavefront-sensing data, calibrates it with a reference star, then adjusts the model to match the telescope’s state during the science observation.

To avoid accidentally removing a planet along with the starlight, the team injects simulated planets into images and tracks their signal-to-noise during optimization, stopping when the signals peak or level off.

The method relies on physics-based modeling and optimization rather than AI, and it outperformed conventional processing on observations of HIP 65426, 51 Eridani and TYC 5899 without requiring changes to JWST.

A standout result was FLIP’s detection of the known giant planet 51 Eridani b in JWST data at 4.3 micrometres, where standard processing had not distinguished it from residual starlight.

The F430M observation is the first image of 51 Eridani b at that wavelength, which includes a carbon-dioxide absorption band that could help future studies assess the planet’s atmospheric metallicity.

In the F410M filter, FLIP increased the signal-to-noise of HIP 65426 b by about 1.6 times and 51 Eridani b by about 2.1 times; it detected 51 Eridani b in F430M at a signal-to-noise ratio of 6.3, where the original analysis reported no detection.

For TYC 5899, whose reference-star observations were taken four months before the science observation, FLIP achieved up to about 15 times better sensitivity than KLIP at 0.5 arcseconds and about five times better at 1 arcsecond.

JWST can currently detect planets roughly the size of Saturn when they orbit far from their stars, while directly imaging Earth-sized worlds would require a telescope 100 to 1,000 times more stable than JWST, according to study lead Rodrigo Ferrer-Chavez.

The researchers hope FLIP will help uncover fainter, cooler and lower-mass planets in existing data and improve studies of known planets; they are reprocessing hundreds of JWST coronagraphic images in search of additional discoveries.