Sinusoidal Single-Pixel Imaging Based on Fourier Positive–Negative Intensity Correlation

Lingtong Meng, Ping Jia, Honghai Shen, Ming-Jie Sun, Dong Yao, Hanyu Wang, Chunhui YanSensors 2020 · DOI

TL;DR

This paper introduces a novel single-pixel imaging (SPI) technique that uses sinusoidal and cosine illumination patterns derived from the Fourier matrix. By correlating "positive" and "negative" intensity measurements, the method reconstructs high-quality images, even under sub-sampling conditions, offering a promising alternative to existing SPI approaches.

What Problem Does This Paper Address?

Traditional digital cameras rely on multi-pixel detectors, where image quality is directly tied to the number of pixels. However, in scenarios with weak signals due to scattering or absorption, or when dealing with specialized imaging needs, multi-pixel detectors become impractical. Single-pixel imaging (SPI) offers a solution by using a single-pixel detector and structured illumination patterns to reconstruct an image over time. The challenge with SPI is the long acquisition time required, which is directly related to the number of illumination patterns used. To reduce this time, researchers have explored orthogonal patterns like Hadamard and Fourier sinusoidal patterns. This paper addresses the need for efficient and high-quality image reconstruction in SPI, specifically by proposing a new method that leverages the properties of Fourier-derived sinusoidal and cosine patterns for improved performance, particularly under reduced sampling rates.

Methods

The core of this work lies in a novel Fourier single-pixel imaging technique that utilizes "Fourier positive-negative intensity correlation." Instead of random patterns, the authors employ sinusoidal and cosine illumination patterns generated from the real and imaginary parts of the discrete Fourier matrix.

  1. Pattern Generation: The complex Fourier matrix is decomposed into its real (cosine) and imaginary (sine) components. These 1D signals are then reshaped into 2D patterns suitable for projection. To ensure they can be projected, the trigonometric function values are shifted to a positive range.
  2. Illumination and Measurement: These sinusoidal and cosine patterns are projected onto the target scene. The light reflected or transmitted is measured by a single-pixel detector, yielding total intensity values.
  3. Positive-Negative Correlation: The technique specifically generates two intermediate images: a "Fourier positive image" and a "Fourier negative image." These are derived from the intensity measurements corresponding to the cosine and sine patterns, respectively. The paper notes that these intermediate images appear as centrosymmetric, overlapping shadows due to missing phase information.
  4. Image Reconstruction: A high-quality final image is reconstructed by applying an intensity correlation algorithm to these positive and negative images. This process effectively acts as a "phase compensation" mechanism, resolving the ambiguities present in the intermediate images.

The authors also introduce metrics for evaluating image quality: Root Mean Square Error (RMSE) for simulations and Signal-to-Noise Ratio (SNR) for experiments.

Datasets

The paper does not explicitly mention or introduce new datasets for training. Instead, it uses:

  • Numerical Simulations: A 128x128 pixel image is used as the target scene for simulations.
  • Experimental Target: A printed USAF 1951 resolution board is used as the physical target in experiments.

Key Insights and Ideas

  • Fourier Matrix as a Source of Orthogonal Patterns: The paper highlights the Fourier matrix not just as a mathematical tool for image transformation but as a direct generator of orthogonal sinusoidal and cosine illumination patterns for SPI.
  • Positive-Negative Correlation for Phase Recovery: The core innovation is the use of intensity correlation between "positive" (cosine-based) and "negative" (sine-based) intermediate images to recover phase information and reconstruct a clear image. This bypasses the need for complex phase-shifting algorithms in some contexts.
  • Potential for Sub-sampling: The technique shows promise for imaging under sub-sampling conditions, meaning fewer illumination patterns are needed to reconstruct a reasonable image. This is crucial for reducing acquisition time.
  • Inherent Limitations of High-Frequency Patterns: The paper acknowledges that using the "natural order" of Fourier matrix patterns can lead to insufficient sampling capacity in high-frequency regions, causing ringing artifacts, especially at low sampling ratios. This suggests that pattern selection or reordering might be important for optimal performance.

Key Results

  • High-Quality Image Reconstruction: Both simulations and experiments successfully reconstructed high-quality images of the target scenes.
  • Feasibility Demonstrated: The proposed technique was validated through both numerical simulations and physical experiments, demonstrating its practical applicability.
  • Performance under Sub-sampling: The paper suggests the technique has potential for sub-sampling, with simulations showing that even at sampling ratios as low as 12.5%, some level of image reconstruction is possible, though with noticeable artifacts.
  • SNR Achieved: In experiments, a Signal-to-Noise Ratio (SNR) of 66.338 dB was achieved for the reconstructed image of the USAF 1951 resolution board.

Limitations and Future Work

  • Ringing Artifacts at Low Sampling Ratios: The authors acknowledge that when using low sampling ratios, especially with high-frequency patterns, ringing effects appear in the reconstructed images. This is attributed to the loss of high-frequency information.
  • Pattern Frequency Selection: The paper notes that the "natural order" of Fourier matrix patterns might not be optimal for low sampling ratios. Future work could involve reordering patterns based on their frequency complexity to improve performance.
  • Illumination Mode: The current setup uses a projector, which has a limited frame rate (6 frames per second in experiments). The authors suggest that using a binary device like a Digital Micromirror Device (DMD) could significantly reduce imaging time, as DMDs are capable of much higher projection speeds.
  • Frequency-Domain Filtering: Future research could explore frequency-domain filtering through pattern selection to ensure both low-frequency components for overall image structure and accurate high-frequency components for detail.

Summary updated

Abstract

Single-pixel imaging techniques extend the time dimension to reconstruct a target scene in the spatial domain based on single-pixel detectors. Structured light illumination modulates the target scene by utilizing multi-pattern projection, and the reflected or transmitted light is measured by a single-pixel detector as total intensity. To reduce the imaging time and capture high-quality images with a single-pixel imaging technique, orthogonal patterns have been used instead of random patterns in recent years. The most representative among them are Hadamard patterns and Fourier sinusoidal patterns. Here, we present an alternative Fourier single-pixel imaging technique that can reconstruct high-quality images with an intensity correlation algorithm using acquired Fourier positive-negative images. We use the Fourier matrix to generate sinusoidal and phase-shifting sinusoid-modulated structural illumination patterns, which correspond to Fourier negative imaging and positive imaging, respectively. The proposed technique can obtain two centrosymmetric images in the intermediate imaging course. A high-quality image is reconstructed by applying intensity correlation to the negative and positive images for phase compensation. We performed simulations and experiments, which obtained high-quality images, demonstrating the feasibility of the methods. The proposed technique has the potential to image under sub-sampling conditions.
BibTeX
@inproceedings{meng2020sinusoidal,
  author = {Lingtong Meng and Ping Jia and Honghai Shen and Ming-Jie Sun and Dong Yao and Hanyu Wang and Chunhui Yan},
  title = {{Sinusoidal Single-Pixel Imaging Based on Fourier Positive–Negative Intensity Correlation}},
  booktitle = {Sensors},
  year = {2020},
  doi = {10.3390/s20061674},
}
Figure