Multi-Object Positioning and Imaging Based on Single-Pixel Imaging Using Binary Patterns

Wenwen Meng, Dongfeng Shi, Wei Yang, Linbin Zha, Yuefeng Zhao, Yingjian Wang2022 · DOI

TL;DR

This paper introduces a novel single-pixel imaging (SPI) technique that uses binary patterns for both locating and imaging multiple objects simultaneously. By leveraging the high modulation speed of Digital Micromirror Devices (DMDs) with binary patterns, the method significantly reduces the number of required measurements, achieving over 85% reduction compared to traditional SPI methods.

What Problem Does This Paper Address?

Single-pixel imaging (SPI) is a promising imaging technology that uses a single-pixel detector to reconstruct images by projecting a series of known illumination patterns onto a scene. A key challenge in SPI, especially for applications requiring high frame rates or high resolutions, is the large number of patterns needed for image reconstruction. This directly impacts the acquisition time and overall efficiency. While Digital Micromirror Devices (DMDs) offer high modulation frequencies for binary patterns, many existing SPI techniques that aim for efficiency, like Fourier SPI, rely on grayscale patterns, which are modulated at much lower frequencies by DMDs. This mismatch limits the speed and applicability of DMD-based SPI systems for real-time imaging of multiple objects. The paper addresses this by proposing a method that utilizes binary patterns for both object localization and subsequent imaging, thereby fully exploiting the capabilities of DMD modulators.

Methods

The proposed method combines two key techniques: Radon SPI for multi-object positioning and Hadamard SPI for image reconstruction.

  • Multi-Object Positioning (Radon SPI): The core idea is to project binary patterns that effectively sample the scene's Radon transform. By using a series of binary patterns at different angles, the system generates projection curves for the scene. These curves are then processed to identify and locate multiple objects within the scene. This approach differs from traditional methods that might require scanning the entire scene or using grayscale patterns for localization. The paper highlights that this binary pattern-based approach is well-suited for DMD modulators.
  • Image Reconstruction (Hadamard SPI): Once the objects are localized and their areas are defined, Hadamard SPI is employed for imaging. Instead of projecting patterns over the entire scene, patterns are generated and projected only onto the identified object areas. This "area-positioning" approach significantly reduces the number of measurements needed for reconstruction, as invalid background areas are discarded. The method then reconstructs the image of each object individually and combines them to form the final scene image.

Datasets

Not discussed in the paper. The experiments were conducted using a custom-built experimental system and simulated data.

Key Insights and Ideas

  • Binary Patterns for Both Positioning and Imaging: The most significant insight is the unified use of binary patterns for both locating objects and then imaging them. This directly addresses the speed limitations of DMD modulators when dealing with grayscale patterns.
  • Exploiting DMD Capabilities: The method is specifically designed to leverage the high-speed binary modulation capabilities of DMDs, making it highly relevant for current SPI hardware.
  • Area-Positioning Efficiency: By first precisely locating objects and then imaging only those specific areas, the method drastically reduces the number of required measurements, leading to significant improvements in imaging efficiency.
  • Reduced Pattern Count: The paper demonstrates a substantial reduction in the number of patterns needed, achieving over 85% reduction compared to traditional methods, which is crucial for real-time applications.

Key Results

  • Reduced Pattern Requirements: The proposed method reduced the number of required patterns by more than 85% compared to traditional SPI methods. For some cases with fewer objects, the reduction reached approximately 93%.
  • Accurate Positioning: Experimental results showed successful positioning of multiple objects (up to five) with a relatively small number of projection angles and patterns.
  • High-Quality Imaging: The reconstructed images, even under challenging conditions (e.g., different contrasts and SNRs), demonstrated good quality, with RMSE values comparable to or better than traditional methods, especially at lower spatial sampling rates.
  • Spatial SR Improvement: The spatial sampling rate (SR) for positioning was found to be very small (e.g., ~0.88%), and the overall SR for imaging was also significantly reduced by discarding background areas.

Limitations and Future Work

  • Low Contrast Objects: The paper acknowledges that the algorithm's performance might be challenged under low-contrast conditions, and improving applicability in such scenarios is a focus for future work.
  • Complex Arrangements: While the method can handle multiple objects, the number of projection angles required for positioning can increase with object complexity and arrangement. Future work could explore more advanced algorithms to optimize angle selection for complex scenes.
  • Further Efficiency Improvements: The authors suggest combining their algorithms with other techniques (e.g., [28-30]) to further enhance the method's efficiency.

Summary updated

BibTeX
@misc{meng2022multi,
  author = {Wenwen Meng and Dongfeng Shi and Wei Yang and Linbin Zha and Yuefeng Zhao and Yingjian Wang},
  title = {{Multi-Object Positioning and Imaging Based on Single-Pixel Imaging Using Binary Patterns}},
  year = {2022},
  doi = {10.3390/s22093211},
}
Figure