Reconstructing compound objects by quantum imaging with higher-order correlation functions

Alexander Mikhalychev, Bänz Bessire, Ilya Karuseichyk, Anton Sakovich, Manuel Unternährer, Dmitry Lyakhov, Dominik L. Michels, André Stefanov, D. MogilevtsevarXiv 2019 · DOI · arXiv

Abstract

Quantum imaging has a potential of enhancing precision of the object reconstruction by using quantum correlations of the imaging field. This is especially important for imaging requiring low-intensity fields up to the level of few-photons. However, quantum imaging generally leads to nonlinear estimation problems. The complexity of these problems rapidly increases with the number of parameters describing the object. We suggest a way to drastically reduce the complexity for a wide class of problems. The key point of our approach is connecting the features of the Fisher information with the parametric locality of the problem, and building the efficient iterative inference scheme reconstructing only a subset of the whole set of parameters in each step. This iterative scheme is linear on the total number of parameters. This scheme is applied to quantum near-field imaging, the inference procedure is developed resulting in super-resolving reconstruction of grey compound transmission objects. The functionality of the method is demonstrated with experimental data obtained by measurements of higher-order correlation functions for imaging with entangled twin-photons and pseudo-thermal light sources. By analyzing the informational content of the measurement, it becomes possible to predict the existence of optimal photon correlations providing for the best image resolution in the super-resolution regime. This prediction is experimentally confirmed. It is also shown how an estimation bias stemming from image features may drastically improve the resolution.

Introduction

Quantum imaging leverages the unique correlations between photons to reconstruct objects with precision that can surpass classical limits, a capability particularly vital for low-light applications where sample damage must be minimized. However, a significant hurdle remains: the mathematical reconstruction of multi-pixel objects from quantum measurements is inherently non-linear.

In classical imaging, a single pixel typically maps to a single point on an object. In contrast, quantum imaging often involves measuring how groups of photons interact with an object simultaneously. As the desired resolution increases, the number of parameters to estimate grows exponentially, leading to a "curse of dimensionality" that makes high-resolution reconstruction computationally prohibitive. This research introduces a novel framework to overcome this complexity, enabling practical super-resolution for complex, compound objects.

Theoretical Framework: Parametric Locality and the Sliding Window Method

The researchers' primary innovation is the Sliding Window Method (SWM). Instead of attempting to solve for every pixel in a large image simultaneously, the SWM processes a small "window" of pixels at a time. It reconstructs a local subset of parameters, shifts the window, and repeats the process iteratively. This approach transforms a massive, non-linear optimization problem into a series of manageable, smaller tasks, reducing computational complexity to a linear scale ($O(M)$) relative to the total number of pixels ($M$).

The Sliding Window Method (SWM) and Fisher Information Matrix (FIM) analysis. (a-b) Banded structure of the FIM indicating parametric locality. (f) Schematic of the iterative SWM process. (h) Reconstructed 1D object profile compared to ground truth.

The theoretical justification for this method lies in Fisher Information Matrix (FIM) analysis. The FIM quantifies how much information a specific measurement provides about a parameter. The authors demonstrated that for a wide class of imaging problems, the FIM is "banded" or diagonally dominant. This property, termed parametric locality, implies that a measurement at a specific point is primarily influenced by object features in its immediate vicinity. This locality is what allows the SWM to function without losing global accuracy.

Experimental Validation

The method was tested using higher-order correlation functions ($G^{(2)}$ and $G^{(3)}$), which analyze coincidences between two or three photons to extract sub-wavelength details. The study utilized two distinct light sources:

  1. Pseudo-thermal light: Created by passing a laser through a rotating ground-glass disk to generate speckle patterns.
  2. Entangled photons: Produced via Spontaneous Parametric Down-Conversion (SPDC) in a nonlinear crystal.

Test objects included standard USAF resolution test charts, specifically a negative "5" digit and 3-slit patterns, designed to challenge the system's ability to resolve features smaller than the classical Rayleigh limit.

Key Results and Insights

The SWM successfully achieved super-resolution—imaging beyond the theoretical diffraction limit of the optical system—across both 1D and 2D objects. The experimental data confirmed that the method is robust and applicable to various types of quantum and correlation-based imaging sources.

Experimental 2D reconstruction of a digit "5". (a) Raw experimental data showing significant blurring. (b) Reconstructed image using the SWM, demonstrating super-resolution. (d) Cross-section comparison between the reconstruction (blue) and ground truth (dashed black).

Beyond computational efficiency, the study revealed two critical insights:

  • The Correlation "Sweet Spot": While it might seem that tighter photon correlations always lead to better resolution, the authors found an optimal correlation width. If correlations are too narrow, the "neighborhood" information required for the SWM to bridge local windows is lost, degrading the overall reconstruction.
  • Beneficial Bias: In statistics, bias is usually avoided. However, the researchers showed that incorporating physical constraints—such as knowing an object's transparency is between 0 and 1 or that it is likely binary—can act as a "beneficial bias" that pushes resolution significantly further than unbiased estimators.

Conclusion and Future Outlook

By exploiting parametric locality, the Sliding Window Method makes high-resolution quantum imaging computationally feasible for the first time. While the method currently relies on the banded nature of the FIM—common in near-field imaging—future work could adapt this framework for other geometries, such as far-field ghost imaging. Additionally, the principles developed here hold promise for quantum state tomography and medical diagnostics, where high-precision imaging with minimal light intensity is paramount.

Summary updated

Abstract

Quantum imaging has a potential of enhancing precision of the object reconstruction by using quantum correlations of the imaging field. This is especially important for imaging requiring low-intensity fields up to the level of few-photons. However, quantum imaging generally leads to nonlinear estimation problems. The complexity of these problems rapidly increases with the number of parameters describing the object. We suggest a way to drastically reduce the complexity for a wide class of problems. The key point of our approach is connecting the features of the Fisher information with the parametric locality of the problem, and building the efficient iterative inference scheme reconstructing only a subset of the whole set of parameters in each step. This iterative scheme is linear on the total number of parameters. This scheme is applied to quantum near-field imaging, the inference procedure is developed resulting in super-resolving reconstruction of grey compound transmission objects. The functionality of the method is demonstrated with experimental data obtained by measurements of higher-order correlation functions for imaging with entangled twin-photons and pseudo-thermal light sources. By analyzing the informational content of the measurement, it becomes possible to predict the existence of optimal photon correlations providing for the best image resolution in the super-resolution regime. This prediction is experimentally confirmed. It is also shown how an estimation bias stemming from image features may drastically improve the resolution.
BibTeX
@misc{mikhalychev2019reconstructing,
  author = {Alexander Mikhalychev and Bänz Bessire and Ilya Karuseichyk and Anton Sakovich and Manuel Unternährer and Dmitry Lyakhov and Dominik L. Michels and André Stefanov and D. Mogilevtsev},
  title = {{Reconstructing compound objects by quantum imaging with higher-order correlation functions}},
  year = {2019},
  eprint = {1908.07461},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1908.07461},
}
Figure