Comparing algorithms for reconstructing digital off-axis Fresnel holograms

Michael Liebling, Michaël UnserProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2005 · DOI

TL;DR

This paper tackles the challenge of choosing the best algorithm for reconstructing 3D information from digital off-axis Fresnel holograms. By simulating the hologram acquisition process and comparing four common reconstruction methods against a known "ground truth," the authors provide valuable guidelines for researchers and engineers, highlighting how algorithm performance varies with experimental parameters like distance and angle.

What Problem Does This Paper Address?

Digital holography offers a powerful way to capture 3D information about an object in a single 2D image (a hologram). This is achieved by recording the interference pattern between light scattered by the object and a reference beam. Reconstructing the original 3D scene from this hologram, however, requires sophisticated algorithms. The "off-axis" geometry, where the reference beam is at an angle to the object beam, is particularly useful because it separates the desired information from unwanted artifacts in the frequency domain, allowing for single-shot acquisition. However, a variety of reconstruction algorithms exist, each with its own strengths and weaknesses depending on the specific experimental setup and the desired outcome (e.g., accurate amplitude or phase retrieval). This paper addresses the need for a systematic comparison of these algorithms to guide users in selecting the most appropriate one for their specific application, considering factors like reconstruction quality and computational complexity.

Methods

The authors employ a simulation-based methodology to rigorously compare four digital hologram reconstruction algorithms:

  • Plain Fresnel Transform: A direct numerical simulation of the physical reconstruction process.
  • Fourier Filtering: This method uses a band-pass filter in the frequency domain to isolate the desired diffraction order and remove artifacts from the zero and twin images.
  • Non-linear Fresnelet Approximation: This approach leverages the spatial and frequency separation of diffraction orders in the "Fresnelet domain" to selectively suppress unwanted components.
  • Least-Squares Estimation (LSE) and Fresnel Transform: A two-step method that first retrieves amplitude and phase in the acquisition plane using LSE and then propagates the wave to adjust focus.

The core of their comparison lies in simulating the hologram acquisition process. This allows them to generate holograms with a known "ground truth" object wave, which is essential for quantitative evaluation. They then reconstruct these simulated holograms using the four algorithms and compare the results to the ground truth using signal-to-noise ratio (SNR) metrics for overall quality, amplitude, and phase.

Datasets

The paper does not use pre-existing datasets. Instead, it introduces a novel benchmark by simulating the hologram acquisition process. This simulation allows for the generation of a controlled "gold standard" wave-front and corresponding holograms under various experimental conditions (object-hologram distance and reference wave angle). This approach provides a unique advantage by offering a known ground truth, which is often difficult to obtain with real experimental data.

Key Insights and Ideas

  • No single algorithm reigns supreme: The performance of each reconstruction algorithm is highly dependent on the experimental parameters, particularly the distance between the object and the hologram plane, and the angle between the object and reference waves.
  • Distance matters significantly: As the distance between the object and the hologram increases, reconstruction quality generally decreases. This is attributed to the limited pixel count of the CCD camera, which can miss information that spreads beyond its boundaries.
  • Angle can compensate for distance (to a point): A larger angle between the object and reference waves can sometimes compensate for a shorter object-hologram distance, improving reconstruction quality. However, there's an upper limit to the angle dictated by the camera's sampling rate.
  • LSE excels in specific scenarios: The Least-Squares Estimation (LSE) method shows superior performance, especially for low angles, due to its local nature, making it less susceptible to boundary artifacts.
  • Fresnelet approximation balances trade-offs: The Fresnelet approximation performs well when a compromise between spatial and frequency filtering is achievable, offering good amplitude reconstructions.
  • Simulation is key for quantitative comparison: The paper highlights the power of simulation for obtaining quantitative comparisons of algorithms, as it provides a known ground truth that is often elusive in real-world experiments.

Key Results

The paper presents extensive quantitative results in the form of SNR plots (Figures 3, 5, and 6) and qualitative reconstructions (Figures 4, 7, and 8).

  • SNR trends: Generally, SNR values increase with increasing object-hologram distance (up to a point) and increasing angle between object and reference waves.
  • Algorithm performance variations:
    • For low angles, LSE generally outperforms other methods.
  • For higher angles, Fourier filtering can yield better amplitude results, especially when considering only the central part of the image.
  • The Fresnelet approximation shows advantages when there's a moderate overlap between diffraction orders.
  • When the distance is zero, only Fourier filtering and LSE can retrieve meaningful information.
  • Qualitative examples: The figures showcase reconstructed amplitudes and phases, visually demonstrating the differences in quality and artifact presence among the algorithms under various conditions. For instance, Figure 4 shows that for d=0.05m and Kx=Ky=-0.6, LSE and Fourier Filtering provide significantly better phase reconstructions than the Plain Fresnel Transform.

Limitations and Future Work

  • Experimental validation: The authors acknowledge that while their simulation-based approach provides valuable quantitative insights, it does not fully replace experimental validation. Real-world experiments can introduce additional complexities not captured in simulations.
  • Parameter space: The study focuses on the influence of object-hologram distance and reference wave angle. The authors suggest that their methodology is suitable for examining a wider range of parameters, such as illumination coherence, noise, and non-ideal optical components.
  • Algorithm evolution: The paper notes that the field of digital holography is rapidly evolving, with new algorithms and hardware advancements. The authors advocate for flexible algorithms that can adapt to these changes and for re-evaluating previously discarded methods in light of modern computing power.
  • Future computational power: The authors anticipate that future advancements in computing power and camera technology will continue to influence the choice and performance of reconstruction algorithms.

Summary updated

Abstract

Three-dimensional information about an object, such as its depth, may be captured and stored digitally in a single, two-dimensional, real- valued hologram acquired in an off-axis geometry. Digital reconstruction of the hologram permits the quantitative retrieval of depth data and object position, or allows post-acquisition focusing on selected scenes. Over the past few decades, a number of reconstruction algorithms have been proposed to perform this task in various experimental conditions and for different purposes (metrology, imaging, etc.). Here, we aim at providing guidelines for deciding which algorithm to apply to a given problem. We evaluate reconstruction procedures based on criteria such as reconstruction quality and computational complexity. We propose a simulation procedure of the acquisition process, that allows us to compare a large body of experimental situations and, because the ground truth is known, achieve quantitative comparison.
BibTeX
@inproceedings{liebling2005comparing,
  author = {Michael Liebling and Michaël Unser},
  title = {{Comparing algorithms for reconstructing digital off-axis Fresnel holograms}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2005},
  doi = {10.1117/12.631039},
}
Figure