PnP-ADMM based efficient hardware implementation of a flexible denoising single-pixel imaging system.

Area of Science:

• Computational Imaging
• Hardware Acceleration
• Image Reconstruction

Background:

• Single-pixel imaging (SPI) offers high sensitivity and interference resistance but faces challenges in real-time noise robustness.
• Existing SPI systems require improvements for practical applications in noisy environments.

Purpose of the Study:

• To develop a hardware-amenable algorithm and system for robust real-time single-pixel imaging.
• To enhance image recovery in SPI systems by integrating denoising operators with a PnP-ADMM algorithm on an FPGA platform.

Main Methods:

• A hardware-amenable Plug-and-Play Alternating Direction Method of Multipliers (PnP-ADMM) algorithm and its IP core were developed.
• A dedicated computing system utilizing FPGA technology was designed for PnP-ADMM implementation.
• Different denoising operators (proximal, soft-thresholding, TV) were evaluated for image recovery.

Main Results:

• The PnP-ADMM algorithm achieved improved Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) under Gaussian noise.
• FPGA implementation drastically reduced image reconstruction time (up to 184.1x faster than ARM processors).
• The system demonstrated a resolution of 0.445-0.5 lp/mm at a 25% sampling rate, outperforming existing methods.

Conclusions:

• The developed FPGA-based PnP-ADMM SPI system offers superior real-time noise robustness and image quality.
• The compact and efficient system design expands SPI applications into IoT and outdoor settings.