Features and Properties of Single-Pixel Imaging Using Speckle Patterns Generated by Multi-Core Fiber

Ryuta Yamaguchi, Kanami Ikeda, Osanori Koyama, Makoto YamadaIEEE Access 2023 · DOI

Introduction

Single-pixel imaging (SPI) has emerged as a robust alternative to traditional multi-pixel camera sensors, particularly in challenging spectral domains such as X-ray, infrared, and terahertz imaging. By using a single detector to capture light intensity from a series of structured illumination patterns, SPI can reconstruct high-quality images. However, conventional SPI systems rely on Spatial Light Modulators (SLMs) to generate these patterns. SLMs are often bulky, expensive, and limited to kilohertz frame rates, which constrains the miniaturization and speed of SPI systems.

To overcome these limitations, researchers are investigating the use of optical fibers. While multimode fibers (MMFs) are sensitive to environmental perturbations like bending, Multi-Core Fibers (MCFs) offer a more stable platform. This study explores the feasibility of using speckle patterns generated by MCFs for high-speed, ultra-compact SPI systems.

Methodology and Simulation Design

The researchers conducted numerical simulations to evaluate the performance of an MCF-based SPI system. The core of the investigation focused on how the physical properties of the fiber—specifically the number of cores—interact with different image reconstruction algorithms.

Pattern Generation

The study simulated MCFs with 7, 14, and 21 cores. By applying random phase shifts to the light within each core, unique speckle patterns were generated via Fraunhofer diffraction at the fiber's end face. These patterns serve as the structured illumination required for SPI.

Reconstruction Algorithms

Four mathematical approaches were compared to determine the most effective method for translating single-pixel measurements back into spatial images:

  • CS-DCT: Compressive sensing utilizing the Discrete Cosine Transform.
  • TVCS (Total Variation Compressive Sensing): An approach focused on preserving edges and minimizing noise.
  • LRCS (Low-Rank Constraint Sensing): A method that exploits the self-similarity inherent in many images.
  • IC Method: An iterative compressive method employing Landweber regularization and guided filtering.

Experimental Datasets

To test the system's versatility, three primary target objects of varying complexity were used in the simulations. These included a simple geometric cross, a handwritten digit "0" from the MNIST dataset, and the text string "OMPU."

Primary test objects used for evaluation: (a) Object 1 (low complexity cross), (b) Object 2 (medium complexity MNIST digit), and (c) Object 3 (high complexity text).

Key Insights and Results

The Impact of Core Count

A critical finding was the relationship between the number of fiber cores and the "rank" or information-carrying capacity of the generated patterns. The 7-core MCF proved insufficient for most imaging tasks due to its low rank. However, increasing the count to 14 or 21 cores significantly improved the variety of speckle patterns, enabling the reconstruction of more complex objects. Interestingly, for simple objects, the 14-core and 21-core fibers performed similarly, suggesting a point of diminishing returns for low-complexity targets.

Algorithm Robustness and Efficiency

While the Iterative Compressive (IC) method performed well in noise-free environments, TVCS emerged as the most robust algorithm when dealing with measurement noise (tested at 10 dB). Furthermore, the system demonstrated remarkable sampling efficiency; for simple objects, stable image reconstruction was achieved at sampling rates as low as 5–7%.

Complexity and Entropy

The study identified a direct correlation between the entropy of the target object and the required core count. If an object's complexity exceeded a specific threshold (approximately 0.186 in this setup), the 14-core fiber failed to produce a recognizable image, necessitating the 21-core configuration. The 21-core system successfully reconstructed complex grayscale images at a 24% sampling rate.

Conclusion and Future Directions

This research demonstrates that MCFs can effectively replace SLMs in SPI systems, offering a path toward ultra-compact imaging devices capable of reaching gigahertz modulation speeds. While the simulation results are promising, future work must address real-world challenges such as inter-core crosstalk and the physical scaling of cladding diameters as core counts increase. Experimental validation will be essential to account for environmental factors like fiber bending and temperature fluctuations in practical applications.

Summary updated

BibTeX
@article{yamaguchi2023features,
  author = {Ryuta Yamaguchi and Kanami Ikeda and Osanori Koyama and Makoto Yamada},
  title = {{Features and Properties of Single-Pixel Imaging Using Speckle Patterns Generated by Multi-Core Fiber}},
  journal = {IEEE Access},
  year = {2023},
  doi = {10.1109/access.2023.3281481},
}
Figure