Illumination Temporal Fluctuation Suppression for Single-Pixel Imaging

Han Wang, Mingjie Sun, Lailiang Song2023 · DOI

TL;DR

This paper introduces a novel normalization protocol for single-pixel imaging (SPI) that effectively suppresses image noise caused by temporal fluctuations in illumination intensity. By leveraging differential measurements and a simple normalization step, the proposed method significantly improves image quality without requiring additional hardware, making SPI more robust and practical.

What Problem Does This Paper Address?

Single-pixel cameras offer advantages in non-visible imaging and cost-effectiveness compared to traditional cameras with pixel arrays. However, a major limitation is the reduced image quality, often due to noise. SPI systems acquire images by taking a sequence of measurements, making them susceptible to inconsistencies caused by variations in illumination intensity during these measurements. While differential measurement techniques can mitigate some noise, temporal fluctuations, especially at high illuminance levels, remain a significant challenge. This paper aims to address this by proposing a method to fully suppress multiplicative noise arising from illumination fluctuations, thereby enhancing the signal-to-noise ratio (SNR) of reconstructed images.

Methods

The core of the proposed solution is a normalization protocol applied to Differential Hadamard Single-Pixel Imaging (DHSI). DHSI, a common SPI technique, uses complementary patterns derived from Hadamard matrices to reconstruct images. The paper builds upon this by:

  • Differential Measurement: Utilizing pairs of inverse patterns (Pᵢ₊ and Pᵢ₋) and their corresponding measurements (Sᵢ₊ and Sᵢ₋) to form a differential signal (Sᵢ₊ - Sᵢ₋). This inherently suppresses additive noise.
  • Normalization Protocol: The key innovation is the normalization step. The sum of the differential signals from two inverse measurements (Sᵢ₊ + Sᵢ₋) is found to be proportional to the illumination intensity. This sum is then used as a weighting factor to normalize the differential signal before image reconstruction. The formula for the normalized signal is: $$S_{i+}^* = \frac{S_{i+}^*}{(S_{i+}^* + S_{i-}^*)/2}$$ This effectively "normalizes" the signal by the local illumination intensity, compensating for fluctuations.

  • No Additional Hardware: Crucially, this protocol does not require a second detector or reference arm, unlike some existing methods (e.g., Normalized Ghost Imaging - NGI). It only requires a slight increase in computational complexity.

Datasets

The paper uses a modified US Air Force (USAF) resolution test chart for both simulations and experimental evaluations. This standard target allows for quantitative assessment of image quality through metrics like Contrast-to-Noise Ratio (CNR).

Key Insights and Ideas

  • Illumination Fluctuation as Multiplicative Noise: The paper clearly identifies temporal illumination fluctuations as a source of multiplicative noise that significantly degrades SPI image quality.
  • Sum of Differential Signals as an Illuminance Proxy: A crucial insight is that the sum of measurements from complementary patterns (Sᵢ₊ + Sᵢ₋) directly correlates with illumination intensity, providing a readily available proxy for normalization.
  • Hardware-Free Noise Suppression: The proposed normalization protocol achieves significant noise reduction without the need for additional optical components like a reference detector, making it highly practical and cost-effective.
  • Complementary to Differential Measurement: The normalization protocol works synergistically with differential measurements, enhancing the overall robustness of SPI systems.

Key Results

  • Significant CNR Improvement: In simulations and experiments, the proposed Normalized-Differential Hadamard Single-Pixel Imaging (NDHSI) method demonstrated substantial improvements in Contrast-to-Noise Ratio (CNR) compared to standard HSI and DHSI, especially under high illumination fluctuation levels. For instance, with a laser output intensity fluctuation of 27.65%, NDHSI achieved a 112% improvement in CNR over DHSI.
  • Performance Close to Reference Arm Methods: The NDHSI method achieved image quality comparable to methods that utilize a reference arm (RDHSI), which requires additional hardware.
  • Minimal Computational Overhead: The added computational cost for the normalization protocol is described as "a little more calculation," making it efficient.
  • Robustness Across Illumination Levels: The protocol showed effectiveness across various illumination fluctuation levels, from low (0.6%) to high (27.65%).

Limitations and Future Work

  • Frequency Dependence: The paper notes that the improvement from the normalization protocol is affected by the frequency and amplitude of illumination fluctuations. While the protocol is effective, the degree of improvement might vary depending on the specific characteristics of the light source.
  • Generalizability: While the paper focuses on Hadamard-based SPI, the authors suggest that the normalization principle could be applied to other SPI techniques that employ differential operations, indicating potential for broader applicability.
  • Future Directions: The authors highlight the potential of this protocol for non-visible imaging applications where light sources are often more sensitive to environmental factors and prone to fluctuations. Further exploration in infrared and terahertz imaging is suggested.

Summary updated

BibTeX
@misc{wang2023illumination,
  author = {Han Wang and Mingjie Sun and Lailiang Song},
  title = {{Illumination Temporal Fluctuation Suppression for Single-Pixel Imaging}},
  year = {2023},
  doi = {10.3390/s23031478},
}
Figure