Spatial multiplexing reconstruction for Fourier-transform ghost imaging via sparsity constraints

Ruiguo Zhu; Hong Yu; Ronghua Lu; Zhijie Tan; Shensheng Han

doi:10.1364/OE.26.002181

1. Introduction

Ghost Imaging (GI) has raised more and more interest in the past decades due to its special characteristics and wide applications. The first experiment about GI was demonstrated with an entangled source generated by spontaneous parametric down-conversion in 1994 [1]. Bennink and his colleagues performed a coincidence imaging experiment with a classical source in 2002 [2]. And theoretical GI schemes exploiting the correlation of two classical beams obtained by splitting incoherent thermal radiation were proposed in 2004 [3, 4]. Soon the experimental demonstration of GI with pseudo-thermal sources and true thermal lights were reported [5, 6]. After that, lots of relevant results including computational, lensless, differential, high-order and turbulence-free GI have been achieved [7–12] and ghost imaging has found its applications in remote sensing, super-resolution, security inspection and x-ray diffraction etc [13–19].

Although ghost imaging have prospered with visible light, it was not realized in the x-ray regime until recently. X-ray Fourier-transform GI of a complex amplitude sample was experimentally demonstrated with a pseudo-thermal source in 2016 [20]. And the real-space x-ray GI experiment was also performed [21]. Nevertheless, in order to achieve high quality images of samples, a great many of measurements are required in x-ray ghost imaging to calculate the average of ensemble, which may result in low efficiency and radiation damage. Fortunately, ghost imaging via sparsity constraints(GISC) was proposed by combining GI methods with the theory of compressing sensing [14, 16, 18], which provides the approach of high quality ghost imaging with less measurements and has already been applied in three dimensional lidar and multispectral camera [22, 23]. In GISC, data acquisition can be treated as an information encoding process. The sample’s information is encoded by the fluctuation of light fields and can be reconstructed from the sensing equation with prior knowledge. Similar method was proved to be effective in Fourier-transform ghost imaging [20, 24]. However, the scale of the sensing matrix is proportional to the square of the image size theoretically, which makes the reconstruction very time consuming and less applicable to high resolution imaging. More importantly, in Fourier-transform ghost imaging, much more correlated information can be obtained simultaneously by simply using a multipoint detector behind the sample, but existing reconstruction method does not take full advantage of it.

In this paper, a spatial multiplexing reconstruction method for Fourier-transform GISC has been proposed to improve the sampling efficiency and image quality of Fourier-transform ghost imaging. The sensing equation of Fourier-transform GISC is established based on recombination and reutilization of the intensity data, and the scale of the sensing matrix is just proportional to the image size. Relevant experiments were performed, and the results are in agreement with theoretical predictions.

2. Theory and method

Figure 1 shows the scheme of Fourier-transform GISC. The laser illuminates a rotation ground glass to produce pseudo-thermal light. A beam splitter divides the pseudo-thermal light into two paths: in the reference arm, the pseudo-thermal light propagates freely to CCD-1; in the test arm, the light transmits the object and detected by CCD-2. Different from traditional single point detector scheme, the detector in the test arm is a multipoint detector. Thus, intensity correlation at different spatial points can be obtained simultaneously, and the intensity signals can be multiplexed to form a more effective sensing equation.

Fig. 1 Scheme of Fourier-transform ghost imaging via sparsity constrains (GISC). BS is a beam splitter, and the information of the object is retrieved from the sensing equation constructed with the intensities recorded by the two CCDs

Download Full Size | PDF

The light field on the detector plane in the reference arm can be described as:

E_{1} (x_{1}, y_{1}) = \int_{s} E_{0} (x_{0}, y_{0}) h (x_{0}, y_{0}; x_{1}, y_{1}) d x_{0} d y_{0},

where E(x₀, y₀) is the optical field in the source plane, h(x₀, y₀; x₁, y₁) is the free-space transfer function from source to detector in the reference arm. Suppose the area of the detector is large enough, we can treat the source as a phase-conjugation mirror [6, 24, 25]. Thus, the light field in the front plane of the object can be written as:

E_{t} (x, y) = \int_{r e f} E_{1} (x_{1}, y_{1}) h^{*} (x_{1}, y_{1}; x, y) d x_{1} d y_{1},

where h(x₁, y₁; x, y) is the free-space transfer function from source to object in the test arm.

Then, we have the light field on the detector plane of the test arm

E_{2} (x_{2}, y_{2}) = \int_{o b j} E_{t} (x, y) t (x, y) h (x, y; x_{2}, y_{2}) d x d y,

where t(x, y) is the transmittance of the object, h(x, y; x₂, y₂) is the free-space transfer function from object to detector in the test arm.

The free-space transfer function from plane (x, y) to plane (ζ, η) is

h (x, y; ξ, η) = \frac{e^{i k z}}{i λ z} e x p {\frac{i k}{2 z} [{(x - ξ)}^{2} + {(y - η)}^{2}]},

where λ is wavelength, k is the wave vector and z is the propagation distance. Substituting Eqs. (1) and (2) into Eq. (3) and using the transfer function in Eq. (4), when satisfying

d_{1} = d_{21} + d_{22}

, we have

E_{2} (x_{2}, y_{2}) = \frac{\exp [\frac{i k}{2 d_{22}} (x_{2}^{2} + y_{2}^{2})]}{λ^{2} d_{22}^{2}} \int_{r e f} E_{1} (x_{1}, y_{1}) \exp [\frac{i k}{2 d_{22}} (x_{1}^{2} + y_{1}^{2})] T (\frac{x_{1} - x_{2}}{λ d_{22}}, \frac{y_{1} - y_{2}}{λ d_{22}}) d x_{1} d y_{1} .

Here, T is the Fourier transformation of t(x, y).

Ignoring several constants, the intensity distribution at the detector of the test arm [26] is

\begin{array}{l} I (x_{2}, y_{2}) = E_{2} (x_{2}, y_{2}) E_{2}^{*} (x_{2}, y_{2}) \\ \approx \iint_{r e f, r e f^{'}} 〈 E_{1} (x_{1}, y_{1}) E_{1}^{*} ({x^{'}}_{1}, {y^{'}}_{1}) 〉 \exp [\frac{i k (x_{1}^{2} + y_{1}^{2} - {x^{'}}_{1}^{2} - {y^{'}}_{1}^{2})}{2 d_{22}}] T (\frac{x_{1} - x_{2}}{λ d_{22}}, \frac{y_{1} - y_{2}}{λ d_{22}}) T^{*} (\frac{x_{1} - x_{2}}{λ d_{22}}, \frac{y_{1} - y_{2}}{λ d_{22}}) d x_{1} d y_{1} d {x^{'}}_{1} d {y^{'}}_{1} \end{array}

The symbol

〈 \cdot 〉

represents the average of ensemble. The approximation works well when the speckle size is small because the law of large numbers guarantees that the sum of independent speckle fields converges to the average ensemble, which means the deviation term tends to be zero. If we construct the sensing equation according to Eq. (6) as processed in [24], the scale of the sensing matrix is decided by the integral, which means it will be proportional to the square of the image size. The higher the resolution, the larger the matrix will be. For example, an ordinary 1024 × 1024 image obtained with 1000 measurements, the matrix scale is up to 10⁹. To solve such an equation, is it is quite demanding for computation hardware.

In fact, Eq. (6) can be simplified. The light from a thermal source will become partially coherent after free-space transportation, and the intensity distribution of this kind of partially coherent light can be described as the Gaussian laser beam distribution [27]. Then the pseudo-thermal light has the degree of spatial coherence

g_{Q}^{(0)} (x_{1} - x_{1}^{'}, y_{1} - y_{1}^{'}) \exp [\frac{{(x_{1} - x_{1}^{'})}^{2} + {(y_{1} - y_{1}^{'})}^{2}}{8 σ_{L}^{2}}],

and the average light intensity distribution

I_{Q}^{(0)} (x_{1}, y_{1}) = {(σ_{L} / σ_{Q})}^{2} A_{L} \exp (- \frac{x_{1}^{2} + y_{1}^{2}}{σ_{Q}^{2}}]

Here σ_Q is the linear dimension of the source, σ_L is the width of spatial coherence, A_L is a constant. Considering the property of the first order intensity correlation function [27], we have

g_{Q}^{(0)} (x_{1}, y_{1}; {x^{'}}_{1}, {y^{'}}_{1}) = g_{Q}^{(0)} (x_{1} - {x^{'}}_{1}, y_{1} - {y^{'}}_{1}) = \frac{〈 E_{1} (x_{1}, y_{1}) E_{1}^{*} ({x^{'}}_{1}, {y^{'}}_{1}) 〉}{\sqrt{〈 E_{1} (x_{1}, y_{1}) E_{1}^{*} (x_{1}, y_{1}) 〉 〈 E_{1} ({x^{'}}_{1}, {y^{'}}_{1}) E_{1}^{*} ({x^{'}}_{1}, {y^{'}}_{1}) 〉}}

Substituting Eqs. (7)-(9) into Eq. (6), we get

\begin{array}{l} I_{2} (x_{2}, y_{2}) = \iint_{r e f, r e f'} \sqrt{I_{1} (x_{1}, y_{1}) I_{1} ({x^{'}}_{1}, {y^{'}}_{1})} \exp [\frac{i k (x_{1}^{2} + y_{1}^{2} - {x^{'}}_{1}^{2} - {y^{'}}_{1}^{2})}{2 d_{22}}] \exp [- \frac{{(x_{1} - {x^{'}}_{1})}^{2} + {(y_{1} - {y^{'}}_{1})}^{2}}{8 σ_{L}^{2}}] \\ \times T (\frac{x_{1} - x_{2}}{λ d_{22}}, \frac{y_{1} - y_{2}}{λ d_{22}}) T (\frac{{x^{'}}_{1} - x_{2}}{λ d_{22}}, \frac{{y^{'}}_{1} - y_{2}}{λ d_{22}}) d x_{1} d y_{1} d {x^{'}}_{1} d {y^{'}}_{1} \end{array}

If the source is large enough, the width of spatial coherence will be small enough because.

σ_{L} \propto \frac{λ d}{σ_{Q}}

Hence the integral term

\exp [- \frac{{(x_{1} - x_{1}^{'})}^{2} + {(y_{1} - y_{1}^{'})}^{2}}{8 σ_{L}^{2}}

is a sharp impulse function. In mathematics, the delta function can be expressed as the limitation of a special sequence of function, which is

\lim_{n \to \infty} n \exp (- n^{2} π x^{2}) = δ (x)

. Thus, in the case that the width of spatial coherence is infinitely small, the term in the integral will operate as a ‘perfect’ Dirac delta function. In practice, we can adjust the related parameters to make the width of spatial coherence small enough, so this term still can be treated as a Dirac delta function. Similar process can be found in [15]. Using the property of delta function in Eq. (10), we obtain

I_{2} (x_{2}, y_{2}) \propto \int_{r e f} I_{1} (x_{1}, y_{1}) {| T (\frac{(x_{1} - x_{2})}{λ d_{22}}, \frac{(y_{1} - y_{2})}{λ d_{22}}) |}^{2} d x_{1} d y_{1} .

Discretizing Eq. (11), we will have

I_{2}^{q} (x_{2 i}, y_{2 i}) = \sum_{j} I_{1}^{q} (x_{1 j}, y_{1 j}) {| T (\frac{x_{1 j} - x_{2 i}}{λ d_{22}}, \frac{y_{1 j} - y_{2 i}}{λ d_{22}}) |}^{2} Δ x_{1 j} Δ y_{1 j}

where q indicates the q-th measurement in ghost imaging. After m measurements, for the fixed point

P (x_{2 i}, y_{2 j})

on the detector of the test arm, the sensing equation is:

Y_{p} = {\begin{matrix} I_{2}^{1} \\ I_{2}^{2} \\ ⋮ \\ I_{2}^{m} \end{matrix}} = A_{P} X = {\begin{matrix} I_{1}^{1} \\ I_{1}^{2} \\ ⋮ \\ I_{1}^{m} \end{matrix}} X .

Here, Ι₁^q (q = 1, 2 …m) is a row vector formed by the intensities recorded in the reference arm in each measurement, X is a column vector corresponding to the squared modulus of the Fourier transformation of the object’s transmittance.

At present, we just use the intensity information of one point $P (x_{2 i}, y_{2 j})$ on CCD-2, however, the detector of the test arm is a multipoint detector. Assuming the detecting area is large enough, we can choose a window centered at point $P (x_{2}, y_{2})$ as the integral domain, the size of which is determined by the image resolution requirements. Then we have

I_{2} (x_{2}, y_{2}) \propto \int_{w i n d o w} I_{1} (x_{1}, y_{1}) {| T (\frac{x_{1} - x_{2}}{λ d_{22}}, \frac{y_{1} - y_{2})}{λ d_{22}}) |}^{2} d x_{1} d y_{1} .

When the window is shifted corresponding to the change of point

P (x_{2 i}, y_{2 j})

, the signal X will be identical. That is to say, we can concatenate the sensing equation of each different point

P (x_{2}, y_{2})

as following

Y = {\begin{matrix} Y_{P_{1}} \\ Y_{p_{2}} \\ ⋮ \\ Y_{p_{n}} \end{matrix}} = {\begin{matrix} A_{p_{1}} \\ A_{p_{2}} \\ ⋮ \\ A_{p_{n}} \end{matrix}} X = A X .

In this way, we obtain the sensing equation of multiple points (n points here). If

X \in R^{N}

, the scale of the sensing matrix is

A \in R^{m n \times N}

. Compared with the theoretical scale

A \in R^{m' \times N^{2}}

in [24], our method is more practical in computation, which is benefited from the simple form of Eq. (11). Furthermore, in terms of mathematical solution, the required measurements can be cut down. Especially, if n is large enough, m can be very small in practice. Even if N² is reduced to N by conjecture in [24], the

m'

measurements required for reconstruction is much larger than the m measurements required in our method.

There are spatial overlap between the windows of different $P (x_{2}, y_{2})$ , which means the intensity signal at each spatial point on the reference detector is reorganized and reutilized in different sensing matrix A_p. It is important to point out that this spatial multiplexing can further improve sampling efficiency. The mutual coherence of the sensing matrix in our method is smaller than the case of traditional single point method (detailed in Appendix). According to the theory of compressive sensing, the smaller the mutual coherence, the less the required measurements. Therefore, better image quality is foreseeable under the same measurements compared with traditional methods, and Fourier-transform ghost imaging with just a few measurements can be expected.

Using similar methods adopted in other GISC applications [14, 16, 18, 24, 28], the signal X is reconstructed by solving the following optimization

M i n {‖ X ‖}_{1} s . t . Y = A X .

Many algorithms can be used to perform above optimization. In this paper, we resort the orthogonal matching pursuit (OMP) algorithm because it is speedy and easy to implement [29]. The Fourier spectrum of the object is obtained by reshaping the signal X into an image according to Eq. (12). And the spectrum can be transformed to the spatial frequency domain by using the factor

λ d_{22}

in coordinates. The standard hybrid input-output (HIO) retrieval algorithm [30] is adopted in this paper to recover the object’s transmittance in the spatial domain.

3. Experimental results

The scheme of our Fourier-transform GISC experiments is showed in Fig. 1. The pseudo-thermal light was generated by illuminating a rotating ground glass with a 532nm laser source, and the diameter of the laser beam was 5mm. The distances in our experiments were $d_{21} = 32.5 c m$ , $d_{22} = 10 c m$ , $d_{1} = 42.5 c m$ . The pixel size of the CCD cameras was $Δ x_{C C D} = 4.6 μ m$ . The coherence size on the CCD plane was 46μm which was obtained by calculating the autocorrelation of the intensity distributions recorded by CCD-1, and it is consistent with the theoretical value 45.2μm derived according to Van Cittert-Zernike theorem.

In the first experiment, we used a double-slit sample. The width of each slit was 90μm and the distance between the slits was 215μm. Thus, the diffraction peak spacing should be 247.4μm theoretically. Figure 2 shows the Fourier spectrums of the double slits obtained with different reconstruction methods. Figure 2(a) is the result of traditional intensity correlation reconstruction, Fig. 2(b) is the Fourier spectrum retrieved using standard single point reconstruction, and Fig. 2(c) is the Fourier spectrum retrieved using spatial multiplexing reconstruction of 40 points. The measurements to obtain the results in (a)(b)(c) are the same number 800. It is almost unable to recognize the Fourier spectrum in Fig. 2(a), and it is hardly to distinguish the diffraction peaks in Fig. 2(b). While in Fig. 2(c), we see the spectrum of the double slits clearly, and the diffraction peak spacing is 248.4μm, which is close to the theoretical value. The peak spacing varies slightly with different random points, and the results converge when sufficient amount of points are multiplexed. Thus, the method of spatial multiplexing reconstruction dramatically improve the image quality compared with previous two methods. Beyond this, we can achieve high quality Fourier spectrum with fewer measurements by means of utilizing more points in spatial multiplexing as demonstrated in Fig. 2(d). The measurements is only 40, which is much less than the measurements required in those traditional methods. The points used for spatial multiplexing is 1000. It is obviously that the spectrum in Fig. 2(d) is as clear as the result in Fig. 2(c). Therefore, by increasing the number of multiplexed points, we can reduce the number of measurements required in Fourier-transform GISC while maintaining the image quality.

Fig. 2 Fourier spectrums of a double slits obtained with different reconstructions. (a) Traditional intensity correlation reconstruction, (b) standard single point reconstruction, (c)spatial multiplexing reconstruction of 40 points, (d) spatial multiplexing reconstruction of 1000 points. The insets in (a)(b)(c)(d) are the reconstructed Fourier spectrums at the CCD plane, and the curves in (a)(b)(c)(d) are the corresponding cross-section distributions. The measurements for (a)(b)(c) is 800, and the measurements for (d) is 40.

Download Full Size | PDF

In the second experiment, we used a 1.42 × 1.42mm² object with letter ‘GI’ in center. The distance parameters are the same as those in the first experiment. The number of measurement is 1200. Figure 3 shows the result Fourier spectrums reconstructed in different ways and the object recovered from corresponding spectrums. Figure 3(a) and Fig. 3(c) are the results of spatial multiplexing reconstruction of 100 points, while Fig. 3(b) and Fig. 3(d) are the results obtained using standard single point reconstruction. The spatial frequency interval in Fig. 3(a) and Fig. 3(b) is $f_{0} = Δ x_{C C D} / (λ d_{22}) = 86.5 m^{- 1}$ , and the dimension of the Fourier spectrums used in the phase retrieval is 1000 × 1000 pixels, so the pixel size in Fig. 3(c) and Fig. 3(d) is $1 / f_{\max} = 1 / (500 f_{0}) = 23.1 μ m$ . It is obviously that the spectrum of spatial multiplexing reconstruction is better and the recovered letters are more legible.

Fig. 3 Fourier spectrums and recovered objects of different reconstruction methods. (a) Fourier spectrum retrieved using spatial multiplexing reconstruction of 100 points, (b) Fourier spectrum retrieved using standard single point reconstruction, (c) object recovered from (a), (d) object recovered from (b).

Download Full Size | PDF

In order to evaluate the image quality quantitatively, a function indicating the visibility of an image is defined as $δ = \frac{1}{N_{x} N_{y}} \sum_{i, j = 1}^{N_{x} N_{y}} \frac{I {(i, j)}_{\max} - I (i, j)}{I {(i, j)}_{\min} + I (i, j)}$ , where I(i,j) is the intensity at pixel (i,j) and I(i,j)_max is the maximum intensity in the image [31].

Another way to estimate the image quality is the signal-to-noise ratio (SNR). Figure 4 shows the statistical distributions of the Fourier spectrums in Fig. 3. The full width at half maximum (FWHM) presents the level of noise, and the difference between the maximum intensity and the background intensity provides the signal information. The blue curve has a narrower peak and a larger maximum value than the red one, which indicates better performance in SNR. The parameter to describe the performance can be defined as $α = (I_{\max} - I_{b a c k g r o u n d}) / F W H M$ . The results of these two kinds of evaluation are demonstrated in Table 1. It can be found that the spectrum reconstructed via spatial multiplexing has better visibility and SNR, which guarantees the successful recovery of the object in spatial domain.

Fig. 4 Statistical distributions of the Fourier spectrums. The blue line is the curve for the spatial multiplexing reconstruction of 100 points in Fig. 3(a), the red line is the curve for standard single point reconstruction in Fig. 3(b).

Download Full Size | PDF

Table 1. Visibility and SNR of different reconstructions

View Table

4. Conclusion

We propose a spatial multiplexing reconstruction method which can be applied in Fourier-transform ghost imaging to obtain high quality Fourier spectrums of samples. We establish the sensing equation of Fourier-transform GISC by recombining and reutilizing the correlated intensity information of light fields. It is theoretically proved that the scale of the sensing matrix can be reduced dramatically. The column dimension of the sensing matrix in our method is equal to the image size, while previously it is equal to the square of the image size. Theoretical analysis and experimental results show that the sampling efficiency of Fourier-transform ghost imaging is greatly improved by adopting spatial multiplexing. High quality spectrum of a double slits is obtained with only 40 measurements, which is much less than the measurements required by traditional methods. While under the same number of measurements, the Fourier spectrums reconstructed via spatial multiplexing have better visibility and signal-to-noise ratio than traditional reconstructions. A sample’s recovered image in spatial domain with better image quality has also been obtained from its spectrum of spatial multiplexing reconstruction. This technology has significant applications in Fourier-transform ghost imaging, due to its potential for reducing x-ray radiation damage and achieving high quality images in x-ray microscopy, and it can also be used in high-speed remote sensing with visible light.

Appendix

The definition of the mutual coherence of a matrix [32] is

μ (A) = \max_{1 \leq i \neq j \leq n} \frac{| 〈 a_{i}, a_{j} 〉 |}{{‖ a_{i} ‖}_{2} {‖ a_{j} ‖}_{2}},

where a_i is the column vector in matrix

A \in R^{m \times n}

, angel brackets represent the inner product, and

{‖ \cdot ‖}_{2}

represents the l₂-norm. Without loss of generality, we set the column vector a_i as normalized vector.

We use the notation A_s to represent the sensing matrix of single point, because the entries of the sensing matrix are non-negative, we have

μ (A_{s}) = \max_{1 \leq i \neq j \leq n} | 〈 a_{i}, a_{j} 〉 | = \max_{1 \leq i \neq j \leq n} 〈 a_{i}, a_{j} 〉 .

The notation A_m represents the sensing matrix of multiple points, then

μ (A_{m}) = \frac{\max_{1 \leq i \neq j \leq n} \sum_{s = 1}^{P} 〈 a_{i s}, a_{j s} 〉}{P} \leq \frac{\sum_{s = 1}^{P} \max_{1 \leq i \neq j \leq n} 〈 a_{i s}, a_{j s} 〉}{P} = \sum_{s = 1}^{P} \frac{μ (A_{s})}{P},

where 1/P arises as a normalized factor and a_is indicates i-th column of single point’s sensing matrix A_s. The inequality sign is because we change the order of sum function and maximize function. The equation above gives the upper bound of the mutual coherence of a sensing matrix of multiple points. Physically, the mutual coherence of different points should be invariant, then the right-hand side of the equation tends to be the mutual coherence of single point. Therefore, the mutual coherence of the sensing matrix of multiple points is smaller than that of single point.

Funding

National Natural Science Foundation of China (NSFC) No. 11627811; National Key Research and Development Program of China No. 2017YFB0503303.

References and links

1. V. Belinskii and D. N. Klyshko, “Two-photon optics: diffraction, holography, and transformation of two-dimensional signals,” J. Exp. Theor. Phys. 78, 259–262 (1994).

2. R. S. Bennink, S. J. Bentley, and R. W. Boyd, “‘Two-Photon’ coincidence imaging with a classical source,” Phys. Rev. Lett. 89(11), 113601 (2002). [PubMed]

3. E. G. Brambilla, M. Bache, and L. A. Lugiato, “Ghost imaging with thermal light: comparing entanglement and classical correlation,” Phys. Rev. Lett. 93(9), 093602 (2004). [PubMed]

4. J. Cheng and S. Han, “Incoherent coincidence imaging and its applicability in X-ray diffraction,” Phys. Rev. Lett. 92(9), 093903 (2004). [PubMed]

5. D. Zhang, Y. H. Zhai, L. A. Wu, and X. H. Chen, “Correlated two-photon imaging with true thermal light,” Opt. Lett. 30(18), 2354–2356 (2005). [PubMed]

6. A. Valencia, G. Scarcelli, M. D’Angelo, and Y. Shih, “Two-photon imaging with thermal light,” Phys. Rev. Lett. 94(6), 063601 (2005). [PubMed]

7. K. W. C. Chan, M. N. O’Sullivan, and R. W. Boyd, “High-order thermal ghost imaging,” Opt. Lett. 34(21), 3343–3345 (2009). [PubMed]

8. X. H. Chen, Q. Liu, K. H. Luo, and L. A. Wu, “Lensless ghost imaging with true thermal light,” Opt. Lett. 34(5), 695–697 (2009). [PubMed]

9. F. Ferri, D. Magatti, L. A. Lugiato, and A. Gatti, “Differential ghost imaging,” Phys. Rev. Lett. 104(25), 253603 (2010). [PubMed]

10. R. E. Meyers, K. S. Deacon, and Y. Shih, “Turbulence-free ghost imaging,” Appl. Phys. Lett. 98(11), 111115 (2011).

11. H. J. Shapiro, “Computational ghost imaging,” Phys. Rev. A 78(6), 061802 (2008).

12. X. H Chen, I. N. Agafonovei, K. H. Luo, Q. Liu, R. Xian, M. V. Chekhova, and L. A. Wu, “LHigh-visibility, high-order lensless ghost imaging with thermal light,” Opt. Lett. 35(8), 1166–1168 (2007).

13. J. Cheng and S. Han, “Incoherent coincidence imaging and its applicability in X-ray diffraction,” Phys. Rev. Lett. 92(9), 093903 (2004). [PubMed]

14. O. Katz, Y. Bromberg, and Y. Silberberg, “Compressive ghost imaging,” Appl. Phys. Lett. 95(13), 131110 (2009).

15. N. Tian, Q. Guo, A. Wang, D. Xu, and L. Fu, “Fluorescence ghost imaging with pseudothermal light,” Opt. Lett. 36(16), 3302–3304 (2011). [PubMed]

16. W. K. Yu, M. F. Li, X. R. Yao, X. F. Liu, L. A. Wu, and G. J. Zhai, “Adaptive compressive ghost imaging based on wavelet trees and sparse representation,” Opt. Express 22(6), 7133–7144 (2014). [PubMed]

17. M. Zhang, Q. Wei, X. Shen, Y. Liu, H. Liu, J. Cheng, and S. Han, “Lensless Fourier-transform ghost imaging with classical incoherent light,” Phys. Rev. A 75(2), 021803 (2007).

18. C. Zhao, W. Gong, M. Chen, E. Li, H. Wang, W. Xu, and S. Han, “Ghost imaging lidar via sparsity constraints,” Appl. Phys. Lett. 101(14), 141123 (2012).

19. S. Gazit, A. Szameit, Y. C. Eldar, and M. Segev, “Super-resolution and reconstruction of sparse sub-wavelength images,” Opt. Express 17(26), 23920–23946 (2009). [PubMed]

20. H. Yu, R. Lu, S. Han, H. Xie, G. Du, T. Xiao, and D. Zhu, “Fourier-transform ghost imaging with hard X rays,” Phys. Rev. Lett. 117(11), 113901 (2016). [PubMed]

21. D. Pelliccia, A. Rack, M. Scheel, V. Cantelli, and D. M. Paganin, “Experimental x-ray ghost imaging,” Phys. Rev. Lett. 117(11), 113902 (2016). [PubMed]

22. W. Gong, C. Zhao, H. Yu, M. Chen, W. Xu, and S. Han, “Three-dimensional ghost imaging lidar via sparsity constraint,” Sci. Rep. 6, 26133 (2016). [PubMed]

23. Z. Liu, S. Tan, J. Wu, E. Li, X. Shen, and S. Han, “Spectral camera based on ghost imaging via sparsity constraints,” Sci. Rep. 6, 25718 (2016). [PubMed]

24. H. Wang and S. Han, “Coherent ghost imaging based on sparsity constraint without phase-sensitive detection,” EPL 98(2), 24003 (2012).

25. D. Z. Cao, J. Xiong, and K. Wang, “Geometrical optics in correlated imaging systems,” Phys. Rev. A 71(1), 013801 (2005).

26. J. W. Goodman, Statistical Optics, (Wiley Classics Library 2000)

27. E. Collett and E. Wolf, “Is complete spatial coherence necessary for the generation of highly directional light beams?” Opt. Lett. 2(2), 27–29 (1978). [PubMed]

28. H. Yu, E. Li, W. Gong, and S. Han, “Structured image reconstruction for three-dimensional ghost imaging lidar,” Opt. Express 23(11), 14541–14551 (2015). [PubMed]

29. J. A. Tropp and A. C. Gilbert, “Signal recovery from random measurements via orthogonal matching pursuit,” IEEE Trans. Inf. Theory 53(12), 4655–4666 (2007).

30. J. R. Fienup, “Phase retrieval algorithms: a comparison,” Appl. Opt. 21(15), 2758–2769 (1982). [PubMed]

31. W. Gong and S. Han, “A method to improve the visibility of ghost images obtained by thermal light,” Phys. Lett. A 374(8), 1005–1008 (2010).

32. M. F. Duarte and Y. C. Eldar, “Structured compressed sensing: From theory to applications,” IEEE Trans Sign. Process. 59(9), 4053–4085 (2011).

Spatial multiplexing reconstruction for Fourier-transform ghost imaging via sparsity constraints

Abstract

1. Introduction

2. Theory and method

3. Experimental results

4. Conclusion

Appendix

Funding

References and links

Cited By

Figures (4)

Tables (1)

Equations (19)

Optics Express

	δ	α
standard single point reconstruction	0.7368	11.85
spatial multiplexing reconstruction of 100 points	0.9177	60.65