Source-linked AI summary
Off-axis digital hologram reconstruction: some practical considerations
Nicolas Verrier, Michael Atlan
TL;DR
Off-axis intensity hologram reconstruction requires choosing numerical propagation methods suited to recording conditions and object geometry. The paper reviews Fourier, convolution, adjustable-magnification, and Fresnelet approaches, then assesses them experimentally to provide rendering guidelines. It concludes that method choice depends on recording conditions, while adjustable magnification can overcome classical constraints if aliases and replicas are controlled.
Problem
The paper addresses how different numerical reconstruction methods should be selected and applied for off-axis intensity holograms under varying recording conditions.
Method
The paper reviews one- to three-FFT Fourier and convolution methods, adjustable-magnification algorithms, Fresnelet decomposition, and experimental reconstructions of optically acquired holograms.
Results
Experimental assessments provide applicability ranges and show that adjustable-magnification methods can produce the same results across the quadratic lens and Fresnel-Bluestein methods.
Takeaways & Limitations
Method choice should follow recording conditions: 1-FFT suits far, extended objects, whereas convolution approaches suit small objects near the sensor.
Takeaways & Limitations
Adjustable magnification requires care because reconstruction horizons smaller or larger than the object extent can cause aliases or replicas.
Abstract
from arXiv · showhide
Holographic rendering of off-axis intensity digital holograms is discussed. A review of some of the main numerical processing methods, based either on the Fourier transform interpretation of the propagation integral or on its linear system counterpart, is reported. Less common methods such as adjustable magnification reconstruction schemes and Fresnelet decomposition are presented and applied to the digital treatment of off-axis holograms. The influence of experimental parameters on the classical hologram reconstruction methods is assessed, offering guidelines for optimal image rendering regarding the hologram recording conditions.
1. Introduction
The paper introduces off-axis digital holography, reviews its recording and reconstruction context, and outlines a comparison of numerical methods for rendering off-axis intensity holograms.
- Motivation and applications: Digitalization improved reconstruction quality, enabled phase retrieval and hologram processing without optical reconstruction, and supported applications across several imaging domains.Examples include fluid mechanics, biomedical imaging, mechanical vibration analysis, and reconstruction of shifted, tilted, or aberrated data.
- Paper scope: The paper reviews common off-axis reconstruction schemes using one to three Fourier transforms, adjustable magnification, alias and replica suppression, and Fresnelet decomposition.Experimental assessment with optically acquired holograms is used to examine method suitability for targeted applications.
- Digital holography: Digital holography numerically reconstructs the field distribution from an optically recorded hologram measured in a diffraction plane.The recorded interference includes object–reference cross terms that contain amplitude and phase information about the diffracted field.
- Off-axis recording: Off-axis recording separates the autocorrelation, real-image, and twin-image terms in the spatial-frequency domain.The reference and object beams meet at a relative angle α selected to satisfy the sampling theorem.
3. Digital hologram reconstruction
Digital hologram reconstruction refocuses the recorded hologram by numerically propagating the optical field to a reconstruction plane. Efficient implementations divide the methods into Fourier-based and convolution-based families according to recording geometry and object extent.
- Propagation formulation: Digital reconstruction performs a posteriori refocusing by calculating backward light propagation from the hologram to the reconstruction plane.This is equivalent to placing the recorded hologram back into the reference beam, which then acts as the reconstruction beam.
- Propagation formulation: The Fresnel transform provides the propagation relationship used throughout the paper for reconstructing off-axis intensity holograms.Its discrete formulation is developed in one dimension because the variables are separable, with straightforward extension to two dimensions.
- Discrete implementation: The reconstruction coordinates are sampled at p∆ξ in the reconstruction plane, while n∆x samples the CCD plane and N denotes the number of sampling points.These sampling variables define the discrete propagation calculation.
- Method families: Single-FFT Fourier approaches suit extended objects far from the sensor, whereas two- or three-FFT convolution methods suit small objects recorded near it.Alternative methods are considered when adjustable magnification or advanced filtering is required.
3.A. Single-FFT method
The single-FFT method efficiently evaluates the discrete Fresnel transform, but its reconstruction sampling is tied to the recording distance, wavelength, sensor pitch, and array size.
- Sampling relation: FFT implementation relates the reconstruction-plane pitch ∆ξ to the CCD-plane pitch ∆x through the discrete Fresnel-transform sampling condition.This relation enables efficient evaluation of the propagation equation using a fast Fourier transform.
- Intrinsic magnification: The intrinsic single-FFT magnification is γ0 = λz/(N∆x^2), equivalently γ = ∆ξ/∆x.Thus the reconstructed horizon-to-sensor extension ratio is closely linked to reconstruction distance.
3.B. Convolution based approaches
Convolution-based reconstruction treats hologram propagation as a linear-system convolution with the Fresnel impulse response and computes it efficiently in the Fourier domain. These approaches provide unitary magnification.
- Linear-system formulation: Convolution-based reconstruction models propagation as the spatial convolution of the hologram with the Fresnel impulse response h_z.This linear-system interpretation forms the basis of the convolution approaches.
- Magnification: Convolution approaches impose unitary magnification, with reconstruction-plane pitch ∆ξ equal to CCD-plane pitch ∆x.The resulting sampling relation is ∆ξ = ∆x.
- Implementation: The convolution product can be computed efficiently in the Fourier domain using fast Fourier transforms and their inverse.The implementation uses the Fourier transform F and inverse transform F−1.
- Angular-spectrum propagation: Angular-spectrum propagation reconstructs the hologram by applying an angular-spectrum transfer function in the spatial-frequency domain.The spatial frequency variable is denoted by u.
- Angular-spectrum propagation: Figure 2 compares angular acceptance for 1-FFT reconstruction and convolution approaches using solid and dashed curves, respectively.The comparison concerns the angular acceptance of the digital holographic reconstruction process.
3.C. Algorithms with adjustable magnification
Adjustable-magnification reconstruction addresses the fixed magnification imposed by single-FFT and convolution methods. The paper reviews alternative schemes based on padding, multistep processing, convolution, and 1-FFT implementations.
- Motivation: Single-FFT magnification depends on recording wavelength and distance, whereas convolution methods impose unitary magnification.Neither approach permits direct adjustment of the reconstructed hologram’s magnification.
- Motivation: Adjustable magnification makes the reconstruction horizon independent of hologram recording parameters, benefiting multi-wavelength holography.
- Existing approaches: Prior approaches use zero-padding or a two-step sequence of 1-FFT reconstructions to control magnification.Zero-padding can increase computational load, despite good results for multiwavelength hologram multiplexing.
- Methods: The paper focuses on two adjustable-magnification algorithms based respectively on convolution and 1-FFT Fresnel-transform implementations.
- Quadratic lens method: The quadratic-lens convolution method pads the hologram, applies a digital spherical wavefront, and modifies the physical reconstruction distance to z′ = γz.The wavefront curvature is defined in terms of system magnification, and reconstruction can use a three-FFT scheme or angular-spectrum propagation.
- Alternative method: A filtered 1-FFT alternative reconstructs the local object field with adjustable magnification and can limit reference-beam distortion effects.The filtering step requires one additional FFT, or two additional FFTs with angular-spectrum implementation.
3.C.2. Fresnel-Bluestein transform
The Fresnel-Bluestein transform rewrites the discrete Fresnel kernel to obtain a convolution-based reconstruction. Compared with the quadratic-lens method, it produces the same adjustable-magnification renderings across the tested magnifications.
- Fresnel-Bluestein transform: The discrete Fresnel transform is recast as a convolution by rewriting the product 2np as n^2 + p^2 − (p − n)^2.
- Fresnel-Bluestein transform: The Fresnel-Bluestein formulation defines the reconstruction through a spatial convolution of two functions, f and g.The magnification γ is independent of hologram recording parameters and can be adjusted at will.
- Experimental comparison: At 0.8γ0, γ0, and 2.5γ0, the Fresnel-Bluestein and quadratic-lens methods yield the same reconstruction results.The comparison uses a USAF resolution-target sector with 228 line pairs·mm−1 at element (7-6).
3.C.3. Aliases and replicas
Adjustable magnification can introduce replicas when γ < γ0 and aliases when γ > γ0, so both effects must be suppressed for high-quality reconstruction. Filtering removes aliases, while cropping and zero-padding remove replicas.
- Fig. 4 compares reconstructions at γ = 0.5 × γ0, γ = γ0, and γ = 4 × γ0, with corresponding replica removal and alias filtering.
- When γ < γ0, replicas appear in the reconstructed image; when γ > γ0, aliases appear and degrade reconstruction quality.
- For γ > γ0, chirp multiplication, low-pass filtering, and multiplication by C∗ remove alias artifacts and produce a high-contrast reconstruction.The filtering window matches the physical extent of the reconstruction horizon at distance z.
- For γ < γ0, cropping the reconstruction to the (γ0/γ) N pixels associated with the object and zero-padding it to the original hologram size removes replicas.
3.D. Fresnelet decomposition
Fresnelet reconstruction decomposes holograms onto Fresnel-transformed wavelet bases. Its multiresolution behavior depends on the Fresnel transform implementation used to construct those bases.
- Fresnelet reconstruction decomposes a digital hologram onto a basis of Fresnel-transformed wavelets.
- Liebling’s construction uses B-splines to generate a semi-orthogonal wavelet-function basis for multiresolution analysis.The construction involves two-scale relations, binomial filtering, and quadrature mirror filters.
- Fig. 6 presents the experimental procedure used to benchmark holographic reconstruction methods.
- Fresnelet bases are calculated by applying the Fresnel transform to the B-spline basis.
- The chosen Fresnel computation scheme determines Fresnelet-transform properties, including whether the reconstruction has adjustable or unitary magnification.
4. Application
Experimental holograms were recorded with an off-axis setup using a 2048 × 2048 CCD and a green 532 nm laser. Reconstructions evaluated USAF targets at three distances selected relative to the sampling condition.
- Off-axis interference holograms were recorded on a 2048 × 2048 pixel CCD with 7.4 µm pixel pitch.
- An inverted USAF target illuminated by a 532 nm green laser was positioned at three distances from the sensor.
- The three distances were chosen so that the reconstructed sampling interval was smaller than, equal to, or greater than the sensor pixel pitch.
- Fig. 7 shows holographic reconstructions of the USAF target located at different distances.
4.A. Classical reconstruction methods
Classical reconstruction methods behave differently according to the sampling relationship between hologram and reconstruction planes. The 1-FFT method can produce aliases when the reconstructed object exceeds the sensor, angular spectrum suits near-sensor objects, and 3-FFT can produce replicas when its impulse response is undersampled.
- 1-FFT reconstruction produces aliases because its Fresnel transform magnifies the object beyond the CCD sensor limits.
- When ∆ξ < ∆x, angular spectrum reconstruction keeps the object within the CCD sensor horizon and is well suited to holograms recorded near the sensor.
- 3-FFT reconstruction embeds the object within the CCD sensor but produces replicas when z < N∆x/λ because its impulse response is ill-sampled.
4.A.3. Reconstruction for ∆ξ > ∆x
For ∆ξ > ∆x, 1-FFT is appropriate for objects larger than the sensor, whereas unit-magnification angular spectrum and 3-FFT approaches can show aliases. Fresnelet reconstruction inherits the behavior of the Fresnel-transform scheme used to construct its basis.
- 4.A.3. Reconstruction for ∆ξ > ∆x: For objects larger than the sensor array, the 1-FFT Fresnel implementation is appropriate, while unitary-magnification angular spectrum and 3-FFT approaches can produce aliases for distant objects.
- 4.A.3. Reconstruction for ∆ξ > ∆x: Classical reconstruction methods are each valid only over limited distance ranges determined by the hologram recording conditions.
- 4.B. Fresnelets: Fresnelet reconstruction depends on the method used to compute the Fresnel transform of the wavelet base, analogous to a multiscale wavelet decomposition on a Fresnel-transformed base.
- 4.B. Fresnelets: For ∆ξ > ∆x, 1-FFT Fresnelet reconstruction gives good results, whereas 3-FFT reconstruction produces aliases.
- 5. Conclusion: The paper overviews off-axis hologram reconstruction methods and investigates the intrinsic properties and limitations of classical schemes.
- 5. Conclusion: Reconstruction choice depends on recording conditions: 1-FFT suits far, extended objects, while convolution approaches suit small objects near the sensor.