The Reconstruction Quality Improvement of Single-Pixel Imaging via Modified Split-Bregman Iteration

Xu Yang, Lu Xu, Long Wu, Chenghua Yang, Wěi Zhāng, Yuehao Liu, Yong ZhangIEEE Access 2020 · DOI

Abstract

A modified Split-Bregman iteration method is proposed to reconstruct the image of target in computational ghost imaging. Different from the conventional Split-Bregman iteration, each reconstruction result of Split-Bregman iteration is denoised by BM3D filter in the proposed, which can help reduce the noise of the final reconstructed result effectively. The method has high quality reconstruction effectiveness and is able to retain the edge and texture detail information well. The numerical simulation and experiment results verify the feasibility of the proposed method. Comparing with other methods qualitatively and quantitatively, our method has high performance in reconstructing image of target.

Introduction

Single-pixel imaging (SPI), also known as computational ghost imaging (GI), offers a unique alternative to traditional multi-pixel cameras. By using a single-point detector and a series of structured light patterns, SPI can reconstruct images in environments where pixelated sensors are impractical or unavailable. However, SPI typically requires a high number of measurements to achieve acceptable image quality, leading to slow acquisition times. While Compressive Sensing (CS) techniques allow for sub-Nyquist sampling, standard reconstruction algorithms like the conventional Split-Bregman iteration often struggle with noise or excessive smoothing, compromising fine details and edges.

Methodology: Modified Split-Bregman Iteration

To address these limitations, this research introduces a modified Split-Bregman iteration that integrates the Block-matching and 3D filtering (BM3D) algorithm directly into the reconstruction loop.

The Split-Bregman iteration is an efficient solver for $L_1$-norm optimization problems, particularly effective for Total Variation (TV) regularization. BM3D is a state-of-the-art denoising filter that leverages non-local self-similarity by grouping similar 2D image patches into 3D stacks for collaborative filtering. By embedding BM3D within each iteration of the Split-Bregman process, the algorithm prevents noise from accumulating and propagating, ensuring that the intermediate results remain clean throughout the convergence process. This hybrid approach combines the structural recovery strengths of optimization with the edge-preserving capabilities of patch-based denoising.

Experimental Setup

The feasibility of the proposed method was tested using both numerical simulations and a physical laboratory setup. The experimental system, as illustrated in the schematic below, utilizes a laser source (1064nm), a Digital Micromirror Device (DMD) for spatial light modulation, and a single-pixel detector.

Schematic diagram of the single-pixel imaging experimental system, including the laser source, DMD, and detection components.

The measurement process involves projecting Hadamard-based speckle patterns onto the target. The reflected light is collected by a receiving antenna and focused onto a single-pixel detector. The resulting data is then processed by a computer using the modified Split-Bregman algorithm to reconstruct the target image.

Results and Discussion

The proposed method was evaluated against several established algorithms, including TVAL3, OMP, and standard Split-Bregman. In simulations using a binary "ZSTU" text image, the algorithm demonstrated high fidelity in reconstructing sharp edges and clear characters.

Reconstruction of the 'ZSTU' target image showing the original image (left) and the result obtained via the proposed modified Split-Bregman method (right).

Quantitative analysis further supports these findings. In tests with grayscale targets like the "cameraman" and "house," the proposed method consistently achieved higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) values compared to competing methods. For instance, in a simulation with 1000 measurements, the method reached a PSNR of 22.648 dB, significantly outperforming TVAL3. The algorithm also showed remarkable resilience to Gaussian white noise, maintaining image clarity where other methods failed.

Conclusion

The modified Split-Bregman iteration significantly enhances the reconstruction quality of single-pixel imaging by incorporating iterative BM3D denoising. While the approach introduces some computational overhead compared to simpler TV-based methods, its superior performance in noise reduction and detail preservation makes it a powerful tool for high-fidelity imaging from sparse data. Future work will focus on optimizing the algorithm for real-time applications and exploring its scalability for larger image resolutions.

Summary updated

Abstract

A modified Split-Bregman iteration method is proposed to reconstruct the image of target in computational ghost imaging. Different from the conventional Split-Bregman iteration, each reconstruction result of Split-Bregman iteration is denoised by BM3D filter in the proposed, which can help reduce the noise of the final reconstructed result effectively. The method has high quality reconstruction effectiveness and is able to retain the edge and texture detail information well. The numerical simulation and experiment results verify the feasibility of the proposed method. Comparing with other methods qualitatively and quantitatively, our method has high performance in reconstructing image of target.
BibTeX
@inproceedings{yang2020reconstruction,
  author = {Xu Yang and Lu Xu and Long Wu and Chenghua Yang and Wěi Zhāng and Yuehao Liu and Yong Zhang},
  title = {{The Reconstruction Quality Improvement of Single-Pixel Imaging via Modified Split-Bregman Iteration}},
  booktitle = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2020.2997678},
}
Figure