Photon-count correlations reveal two pinholes hidden in conventional ghost imaging

What sets the resolution?

Diffraction limits how closely spaced features can be distinguished in conventional optical imaging. In thermal-light ghost imaging, an image is reconstructed from correlations between measurements in two light beams rather than from a directly recorded picture. The width of the spatial correlation determines how finely the system can distinguish features.

Yu Gu and colleagues investigate whether selecting particular photon counts in the two beams can make that correlation narrower. Their paper, submitted to arXiv on 9 November 2026, is a preprint submitted for publication; the supplied source does not establish peer-reviewed publication.

The central result is a one-dimensional image of two pinholes that standard ghost imaging did not distinguish. Separate measurements also show substantial narrowing of spatial and temporal correlations. Those measurements are important, but they are not equivalent to demonstrating the same improvement in a complete image.

Selecting two photons—and none

The experiment uses pseudothermal light: laser light scattered from rotating ground glass to produce thermal-like fluctuations. The laser wavelength is 780 nanometres. A beam splitter directs the light into two arms of a Hanbury Brown–Twiss interferometer, an arrangement for measuring correlations between light-detection events.

The researchers record photon arrival times and calculate correlations offline. Instead of using only the usual intensity correlation, they select events according to the number of photons detected within a time bin. Their principal example compares events with two detected photons in one arm against events with no detected photons in the other.

Here, “zero photons” means a selected detection outcome, not an image obtained without collecting light. Repeated measurements in both arms provide the statistics needed to reconstruct the correlation.

The selected correlation can contain a peak or a dip. A dip means that the chosen detection outcomes occur together less often than their independent probabilities would suggest. Its shape depends on the average detected photon number and the balance between the two arms.

The authors interpret the narrowing as the combination of a correlation peak and an anticorrelation dip with different widths. They connect this interpretation to both classical techniques that reshape an imaging response and quantum descriptions of multiphoton components. The illumination itself remains classical thermal-like light; the demonstration does not require an entangled-photon source.

What the measurements show

The team characterizes sharpness using the full width at half maximum, a measure of the width of a correlation peak or dip relative to its background.

For temporal correlations, the reference coherence time—the characteristic duration of the light’s intensity correlations—was 173 ± 5 microseconds. The narrowest selected correlation measured 14.7 ± 1.5 microseconds, giving the reported 11.8-fold improvement. This is a temporal correlation measurement, not a demonstrated image of a time-varying object.

For spatial correlations, the reference width was 2.54 millimetres. Selecting two-photon versus zero-photon events reduced the measured width to 0.50 millimetres, corresponding to a 5.1-fold improvement.

The imaging test used two pinholes, each 0.1 millimetres across, separated by 2.4 millimetres between their centres. One arm collected light transmitted through the object, while the other was scanned to reconstruct a one-dimensional image from correlations. The standard ghost-imaging reconstruction did not show two distinct pinholes. With the selected photon-count correlation, the two features became distinguishable as the average detected photon number increased.

Sharper correlations have a cost

The authors explicitly describe the reconstructed image quality as poor. They also caution that the largest measured correlation-width improvements are difficult to achieve in imaging.

As the selected correlation narrows, its surrounding background rises. Contributions from nearby positions then affect the reconstruction, creating a trade-off between narrow features and image quality. Unequal light transmission through the two pinholes also produces dips of different depths.

Only one-dimensional spatial imaging is demonstrated. Two-dimensional imaging, temporal imaging and lithography are proposed extensions, not experimental outcomes of this paper. The result therefore supports a specific conclusion: photon-count selection can sharpen thermal-light correlations enough to distinguish an otherwise unresolved test structure, while useful imaging performance remains constrained by the correlation background.