Related Experiment Video
Updated: Aug 15, 2026
11:23
Lensless Fluorescent Microscopy on a Chip
Published on: August 17, 2011
Learnable foveated sampling and dual-prior reconstruction for single-pixel imaging
Optics Express
|August 14, 2026
Summary
This study introduces FDP_Net, a novel deep learning approach for single-pixel imaging (SPI). It enhances image center reconstruction by using foveated sampling and dual-prior network, improving accuracy in SPI microscopy.
Area of Science:
- Optics and Photonics
- Computer Vision
- Machine Learning
Background:
- Single-pixel imaging (SPI) is a compressed sensing technique with potential for low-cost, high-sensitivity imaging.
- Current deep learning SPI methods often use uniform sampling, not leveraging SPI's flexible sampling patterns.
- Biological vision systems offer insights into efficient high-resolution perception.
Purpose of the Study:
- To develop an advanced deep learning framework for single-pixel imaging (SPI) reconstruction.
- To optimize sampling strategies and reconstruction priors for improved image quality, particularly in the central region.
- To validate the proposed method on a real-world SPI microscopy system.
Main Methods:
- Proposed FDP_Net, a deep network integrating learnable foveated sampling and dual-prior reconstruction.
- Implemented a foveal sampling module for adaptive high-resolution center sampling and aggregated peripheral sampling.
- Developed a deep unfolding reconstruction network with sparse prior module (SPM) and denoising prior module (DPM) fusion.
- Introduced a center-weighted differential loss function to prioritize region-of-interest reconstruction quality.
Main Results:
- FDP_Net achieved significantly better reconstruction accuracy in the image center region compared to mainstream methods.
- The foveated sampling strategy effectively increased the sampling rate in the central image area.
- The dual-prior fusion captured comprehensive image texture and structure information.
- The method was successfully validated on a single-photon counting SPI microscopy system after binarizing the sampling matrix.
Conclusions:
- The proposed FDP_Net effectively optimizes foveated sampling and dual-prior reconstruction for SPI.
- This approach enhances reconstruction accuracy, especially in the image center, outperforming existing methods.
- The successful validation on a real SPI microscopy system demonstrates practical applicability.
