Quantum Ghost Imaging Overview

• Quantum ghost imaging is an imaging modality that reconstructs objects via second-order correlations from paired detectors, where one detector never views the object.
• The technique leverages quantum resources such as entangled photon pairs and matter waves to enable high sensitivity in low-light regimes and nonlocal imaging.
• Advanced protocols incorporate state engineering, multiplexed reconstruction, and inverse problem methods to improve resolution, robustness, and imaging speed.

Searching arXiv for current, relevant quantum ghost imaging papers to ground the article in cited literature. arXiv search query: "quantum ghost imaging review phase imaging atoms HOM SPAD memory multiplexed". Quantum ghost imaging is an imaging modality in which an object is reconstructed from correlations between two spatially separated measurement channels rather than from a directly formed image. In its canonical form, one member of a correlated pair interacts with the object and is detected without spatial resolution by a bucket detector, while the partner particle never encounters the object but is measured with spatial resolution; neither arm alone contains a usable image, and the object emerges only in the joint statistics (Tian et al., 16 Jun 2025). Within this general framework, the literature now includes entangled-photon implementations, matter-wave realizations with ultracold atoms, engineered-state protocols based on Hong–Ou–Mandel interference, single-pixel phase imaging, multiplexed low-photon-number reconstruction, hybrid photon–memory and electron–photon architectures, and long-distance fiber-based variants (Khakimov et al., 2016, Bornman et al., 2019, Sephton et al., 2022, Balakin et al., 2018, Mazelanik et al., 2021, Bogdanov et al., 18 Sep 2025, Dong et al., 2015).

1. Defining architecture and historical scope

Quantum ghost imaging originated as an entanglement-based imaging concept and was experimentally verified with entangled photon pairs; a later review places the proposal with Klyshko in 1988 and the first verification with Pittman and coworkers in 1995 (Tian et al., 16 Jun 2025). In the formulation common to the literature, the test or object arm illuminates the sample and is measured by a bucket detector, whereas the reference arm does not interact with the sample but carries the spatial information needed for reconstruction. The image is therefore a second-order object, not a first-order intensity pattern in either arm alone (Tian et al., 16 Jun 2025).

For photonic implementations based on correlated pairs, this logic is expressed through second-order correlation functions. In the atom-ghost-imaging formulation, the standard background expression is

$$G^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \langle \hat{E}_b^{(-)}(\mathbf{r}_b)\hat{E}_i^{(-)}(\mathbf{r}_i)\hat{E}_i^{(+)}(\mathbf{r}_i)\hat{E}_b^{(+)}(\mathbf{r}_b) \rangle,$$

or, in normalized form,

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$

With bucket integration over the object arm, the ghost image is written as

$$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$

which makes explicit that the object transmission $T(\mathbf{r}_b)$ is recovered nonlocally from joint statistics rather than direct spatial readout in the test arm (Khakimov et al., 2016).

A persistent theme in the field is that ghost imaging is not uniquely quantum in the sense of requiring entanglement for all implementations. The multi-wavelength review explicitly distinguishes quantum-entangled-photon GI from classical pseudothermal, computational, compressive, and deep-learning-assisted variants, while preserving the common bucket-plus-reference correlation architecture (Tian et al., 16 Jun 2025). By contrast, papers devoted specifically to quantum ghost imaging retain pair-generated correlations as the operative resource, whether with SPDC photons, Bose-condensed atoms, or other nonclassical sources (Khakimov et al., 2016, Hodgman et al., 2019).

2. Quantum resources, correlations, and state engineering

The best-known quantum resource in ghost imaging is the spatially entangled biphoton. In near-field SPDC language, one paper writes the two-photon state in a discrete position basis as

$$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$

so that both photons are correlated with the same transverse coordinate in the crystal plane (Bornman et al., 2019). Another single-pixel phase-imaging work uses the correlated state

$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$

again emphasizing discrete position correlation in an image-plane basis (Sephton et al., 2022). In both cases, the point is the same: spatial correlation is the substrate on which object information is transferred from one arm to the other.

Quantum ghost imaging is not restricted to photons. The first demonstration with massive particles used correlated metastable helium atoms created in colliding Bose–Einstein condensates via $s$-wave scattering, with back-to-back momentum correlations

$\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$

There, one atom of a correlated pair passes through an effective mask and is bucket-detected, while its partner atom, which never encounters the mask, is spatially resolved in the opposite halo region (Khakimov et al., 2016). A later extension used the same matter-wave platform to realize higher-order ghost imaging up to fifth order, showing that the source supports not only pair correlations but also a hierarchy of higher-order correlations exploitable for imaging (Hodgman et al., 2019).

State engineering can alter the very form of the ghost image. In an SPDC-based experiment with Dove prisms and Hong–Ou–Mandel interference, the biphoton state is decomposed into symmetric and anti-symmetric sectors,

$$|\Psi\rangle = \sum_{\ell} a_\ell \left( |\Psi_\ell^+\rangle \cos(2\ell\theta) + i |\Psi_\ell^-\rangle \sin(2\ell\theta) \right),$$

with

$$|\Psi_\ell^-\rangle = \frac{1}{\sqrt{2}}\left(|\ell\rangle_A|-\ell\rangle_B-|-\ell\rangle_A|\ell\rangle_B\right).$$

Without the beamsplitter, the ghost image is simply rotated by $$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$0; with HOM-selected anti-symmetric states, the reconstructed intensity becomes

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$1

producing a superposition of two oppositely rotated ghost images (Bornman et al., 2019).

Hyper-entanglement extends the same logic to transparent polarization-sensitive phase objects. In that case the source combines EPR-like spatial entanglement with polarization entanglement, and the object acts only on one polarization component: $$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$2 The image is then not an amplitude shadow but a correlation pattern linking the polarization-momentum of the interacting photon to the polarization-position of the non-interacting one (Saxena et al., 2023).

3. Image formation laws and inverse problems

In its simplest form, ghost imaging reconstructs transmission or reflectivity from a second-order correlation map. The matter-wave formulation expresses this as

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$3

and, when the source correlations are anti-collinear, the ghost image becomes the object convolved with the pair-correlation kernel (Khakimov et al., 2016). A closely related form appears in temporal quantum ghost imaging over fiber, where frequency anti-correlation is converted into a position–time correspondence and the coincidence function simplifies to

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$4

so that a temporal histogram becomes a one-dimensional image of object reflectivity (Dong et al., 2015).

For multiplexed quantum ghost imaging with four-frequency entangled light from coupled parametric processes, image formation is more structured. The source arises from

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$5

and the field operators satisfy

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$6

In that setting the imaging condition depends on whether the relevant channel behaves as down-conversion-type or up-conversion-type. For $$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$7,

$$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$8

whereas for $$g^{(2)}(\mathbf{r}_b,\mathbf{r}_i) = \frac{G^{(2)}(\mathbf{r}_b,\mathbf{r}_i)} {\langle \hat{n}_b(\mathbf{r}_b)\rangle \langle \hat{n}_i(\mathbf{r}_i)\rangle }.$$9,

$$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$0

Under these imaging conditions, the bucket-integrated second-order correlations reduce to

$$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$1

with distinct $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$2 for the different frequency channels (Balakin et al., 2018).

That same work reformulates multiplexed GI as a discrete inverse problem: $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$3 where $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$4 is the unknown image, $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$5 stacks the correlator outputs, and $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$6 includes full inter-channel covariance. The optimal linear unbiased reduction operator is

$$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$7

with $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$8 for direct image reconstruction. The method is then strengthened by a Mahalanobis-distance projection onto the feasible set $$I_{\rm GI}(\mathbf{r}_i) \propto \int_{\text{object arm}} d\mathbf{r}_b \, T(\mathbf{r}_b)\, G^{(2)}(\mathbf{r}_b,\mathbf{r}_i),$$9 and by transform-domain thresholding in a sparse basis such as the DCT or Haar basis: $T(\mathbf{r}_b)$0 This reconstruction strategy is explicitly designed for low-photon-number operation and correlated, object-dependent noise (Balakin et al., 2018).

Phase retrieval and projection-based inference form another major branch of the inverse-problem literature. For pure phase objects in single-pixel quantum ghost imaging, the object is expanded in a complete mask basis,

$T(\mathbf{r}_b)$1

and the coincidence probability for cosine-type projection masks

$T(\mathbf{r}_b)$2

becomes

$T(\mathbf{r}_b)$3

A complementary sine mask

$T(\mathbf{r}_b)$4

yields

$T(\mathbf{r}_b)$5

From these two datasets the experiment reconstructs

$T(\mathbf{r}_b)$6

and

$T(\mathbf{r}_b)$7

so that the full phase profile is recovered up to a global offset (Sephton et al., 2022).

4. Experimental platforms and modality expansion

Quantum ghost imaging now spans a broad set of physical platforms. Photon-pair SPDC remains central, but the range of source engineering and detection architectures has widened considerably. A practical speed milestone was reached with infrared quantum ghost imaging based on non-degenerate SPDC and a $T(\mathbf{r}_b)$8 SPAD array operated in a “looking back” timing mode. There the pump was $T(\mathbf{r}_b)$9, the signal wavelength inferred at $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$0, and the idler centered at $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$1. The object was probed in the infrared arm and bucket-detected with an InGaAs detector, while spatially resolved detection remained in the visible on a silicon SPAD array. The system reported recognizable features after $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$2–$$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$3 and a ghost image in under one minute, with a typical observation rate of $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$4 (Gili et al., 2022).

Long-distance distribution has been demonstrated by abandoning spatial correlations in favor of frequency anti-correlation. In that architecture, the biphoton state

$$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$5

is used so that at Alice frequency is mapped to position on the object, while at Bob chromatic dispersion maps frequency to arrival time. The experiment demonstrated temporal quantum ghost imaging over $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$6 of optical fiber (Dong et al., 2015).

Matter-wave realizations expanded the field beyond optics. The first atom-based demonstration used ultracold metastable helium atoms from a Bose–Einstein-condensate collision halo, and a later work extended the same platform to higher-order ghost imaging up to fifth order. In that higher-order regime, the visibility metric

$$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$7

improved when extra conditioning atoms were placed in the bucket arm, while the resolution remained close to the pair correlation length

$$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$8

The best visibility at fixed order $$|\Psi\rangle = \sum_{\mathbf r \in \mathcal S} c(\mathbf r)\, |\mathbf r\rangle_A |\mathbf r\rangle_B,$$9 occurred with one image-port atom and $$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$0 bucket-port atoms (Hodgman et al., 2019).

Hybrid platforms push ghost imaging into new detector and source regimes. A free electron–photon implementation in a transmission electron microscope used electron–cathodoluminescence photon pairs generated at a $$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$1 silicon membrane by a $$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$2 electron beam. The photon arm carried the mask and was bucket-detected by an SPCM, while the electron arm was spatially resolved on a Timepix3 detector. Coincidence selection used

$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$3

with a $$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$4 window, and the reported spatial resolution reached

$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$5

at the TEM sample plane (Bogdanov et al., 18 Sep 2025).

A separate hybrid direction uses a quantum memory rather than a second propagating photon. In a cold-$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$6 emissive memory, a signal photon and a stored spin wave are generated first, and the idler photon is created only after readout of the memory. The resulting real-time ghost-imaging protocol exploits wavevector multiplexing and feedback-controlled readout. Through a single far-field image, the experiment certified Bell nonlocality with

$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$7

thereby linking ghost imaging to hybrid light–matter entanglement rather than only free-space photon pairs (Mazelanik et al., 2021).

5. Performance determinants, low-light regimes, and limitations

Across implementations, image quality is governed by a small number of recurring factors: source correlation width, detector timing and spatial resolution, noise backgrounds, and the structure of the reconstruction algorithm. In atom ghost imaging, the image sharpness is mainly limited by the width of the back-to-back correlation peak and detector resolution; in higher-order atomic GI, resolution remains essentially fixed while visibility improves or degrades depending on how conditioning particles are distributed between bucket and image arms (Khakimov et al., 2016, Hodgman et al., 2019).

Low-light operation has been studied both with explicit photon-number projection and with covariance-aware reconstruction. In computational quantum projection ghost imaging, each DMD micromirror is treated as a binary random variable $$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$8 with

$$\ket{\Phi}=\frac{1}{N}\sum_{j=0}^{N-1}\ket{r_j}_A\ket{r_j}_B,$$9

and the reconstruction conditions on a fixed bucket photon number $s$0. For vacuum projection, $s$1, the visibility becomes

$s$2

and the CNR is

$s$3

Because $s$4, vacuum projection always yields a negative image, and the paper reports that vacuum-state projection is superior to conventional and fast first-photon ghost imaging in low-light illumination (Cao et al., 2021).

The multiplexed low-photon-number reconstruction work gives a different route to the same regime. There the algorithm operates with full noise covariance, not a Poisson-only approximation, and computer experiments considered illumination as low as about $s$5 photon per pixel. Two object classes were studied on $s$6 images with reference pixels three times larger than object pixels. For a two-slit object, Haar sparsity was better matched to sharp boundaries; for a smoother grayscale image, DCT sparsity was more appropriate. The paper concludes that $s$7 is typically a good compromise between denoising and detail retention (Balakin et al., 2018).

Several papers also emphasize robustness to noise that does not carry object information. In the multiplexed frequency-entangled study, ordinary imaging with $s$8 photons per pixel plus $s$9 noise photons was visually less robust than ghost imaging with extra $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$0 or $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$1 noise photons, especially after reduction-plus-sparsity processing. The stated reason is that uncorrelated noise photons are naturally rejected by coincidence or correlation formation unless they enter the finite coincidence window and bandwidth (Balakin et al., 2018). This does not imply immunity to all backgrounds; rather, it identifies the correlation stage itself as a noise discriminator.

Practical limitations remain stringent. The SPAD-array infrared experiment is constrained by its $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$2 format, $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$3 fill factor, TDC nonuniformity, and the dead time of the InGaAs bucket detector (Gili et al., 2022). Temporal fiber ghost imaging required $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$4 per line and suffered from fiber-length drift over $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$5 (Dong et al., 2015). Electron–photon ghost imaging required a $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$6-hour run and $\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$7 coincidences for a single complex mask (Bogdanov et al., 18 Sep 2025). These constraints suggest that, in present hardware, quantum ghost imaging often exchanges detector simplicity or spectral flexibility for acquisition-time and count-rate penalties.

6. Conceptual extensions and relation to the wider GI landscape

The literature increasingly treats quantum ghost imaging not only as a nonlocal imaging technique but as a platform for broader quantum information processing. One recent work makes this explicit by linking ghost imaging to Grover’s search algorithm. There the spatially entangled state

$\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$8

is combined with an object acting as a phase oracle

$\mathbf{k}_1 + \mathbf{k}_2 \approx 0.$9

so that marked pixels receive a $$|\Psi\rangle = \sum_{\ell} a_\ell \left( |\Psi_\ell^+\rangle \cos(2\ell\theta) + i |\Psi_\ell^-\rangle \sin(2\ell\theta) \right),$$0-phase flip. After bucket detection, the idler acts as a search register and projective masks

$$|\Psi\rangle = \sum_{\ell} a_\ell \left( |\Psi_\ell^+\rangle \cos(2\ell\theta) + i |\Psi_\ell^-\rangle \sin(2\ell\theta) \right),$$1

yield probabilities

$$|\Psi\rangle = \sum_{\ell} a_\ell \left( |\Psi_\ell^+\rangle \cos(2\ell\theta) + i |\Psi_\ell^-\rangle \sin(2\ell\theta) \right),$$2

The reconstruction

$$|\Psi\rangle = \sum_{\ell} a_\ell \left( |\Psi_\ell^+\rangle \cos(2\ell\theta) + i |\Psi_\ell^-\rangle \sin(2\ell\theta) \right),$$3

is then interpreted as a Grover-like retrieval of marked spatial basis states (Gounden et al., 15 Aug 2025). This suggests a shift in emphasis from “image formation” alone to operator-level processing of high-dimensional photonic states.

At the same time, the field’s broader evolution has made clear that ghost imaging is now a family of correlation-based methods rather than a single entanglement-specific protocol. The multi-wavelength review describes a progression from quantum-entangled-photon GI to classical pseudothermal, computational, compressive, and deep-learning variants across XUV, X-ray, visible, infrared, THz, and matter-wave regimes (Tian et al., 16 Jun 2025). Within that broader landscape, quantum ghost imaging remains distinguished by its use of genuinely pair-generated or entanglement-enabled resources, but many of its modern technical advances concern issues that also matter outside the strict quantum domain: sparse reconstruction, phase retrieval, multi-wavelength operation, wide-field detection, and robustness in scattering or low-dose conditions.

A recurring misconception is that ghost imaging is defined by one specific source class. The reviewed work does not support that view. Instead, it supports a narrower and more precise statement: quantum ghost imaging denotes those ghost-imaging protocols in which the operative correlations are supplied by nonclassical sources such as SPDC biphotons, collision-halo atom pairs, or hybrid light–matter states, while the defining imaging logic remains the same—object information is transferred through correlations to a detector arm that never directly views the object (Khakimov et al., 2016, Mazelanik et al., 2021, Tian et al., 16 Jun 2025). A plausible implication is that future developments will continue to separate the imaging architecture from the physical carrier, as already seen in photons, atoms, electron–photon pairs, and quantum memories.