Research map / Learning theory and algorithms / Sample complexity
Image Reconstruction from Readout-Multiplexed Single-Photon Detector Arrays
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a probabilistic multiphoton estimator that redistributes counts from ambiguous row-column readout frames of SNSPD arrays, resolving up to four coincident photons, and evaluates it by Monte Carlo simulation.
Problem
Row-column readout multiplexing of single-photon detector arrays (SNSPDs) produces ambiguous frames when multiple photons arrive in one acquisition; conventional methods either discard these frames or introduce bias, limiting usable photon flux and efficiency.
Approach
Model photon events that make row-column readouts ambiguous, then derive an approximate maximum-likelihood estimator that redistributes counts from ambiguous frames to candidate pixels using conditional event probabilities estimated from unambiguous single-photon frames, without spatial priors.
Key results
- The multiphoton estimator improves PSNR by roughly 3 to 4 dB over the naive and single-photon estimators at optimal flux (4-photon estimator 26.74 dB vs naive 22.59 dB on the tested image).
- The multiphoton estimator reaches MSE 0.01 with 25,000 frames versus 100,000 for the single-photon estimator, about a 4x reduction in frames.
- The multiphoton estimator's MSE matches the Cramer-Rao bound over a range of incident photon fluxes.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use the multiphoton estimator instead of the single-photon estimator | row–column readout reconstruction · 32×32 images, optimal mean PPF | PSNR: 3 to 4 dB higher than the NE and the SPE | Helped |
| use the multiphoton estimator instead of the single-photon estimator | row–column readout reconstruction · 32×32 image, target MSE 0.01 | readout frames: 100, 000 -> 25, 000 | Helped |
| use the multiphoton estimator instead of the naive estimator | row–column readout reconstruction · 32×32 arbitrary images | PSNR: around 6 to 11 dB better | Helped |
| use the multiphoton estimator instead of the single-photon estimator | row–column readout reconstruction · 32×32 arbitrary images | PSNR: 4 to 6 dB better | Helped |
| model up to 3-photon coincidences instead of up to 2-photon coincidences (ablation) | multiphoton estimator · 2×2 array, optimal mean PPF | PSNR: 23.83 dB -> 24.60 dB | Helped |
| model up to 4-photon coincidences instead of up to 3-photon coincidences (ablation) | multiphoton estimator · 2×2 array, optimal mean PPF | PSNR: 24.60 dB -> 26.74 dB | Helped |
| increase modeled coincidence order from 2 to 3 photons (ablation) | multiphoton estimator · 2×2 array | optimal mean photons per frame: 1 -> 1.2 | Helped |
| increase modeled coincidence order from 3 to 4 photons (ablation) | multiphoton estimator · 2×2 array | optimal mean photons per frame: 1.2 -> 1.4 | Helped |
| increase incident flux from 3 to 4 mean photons per frame (ablation) | single-photon estimator · 32×32 image, row (b) to row (e) | PSNR: 22.43 dB -> 19.59 dB | Hurt |
| increase incident flux from 3 to 4 mean photons per frame (ablation) | multiphoton estimator · 32×32 image, row (b) to row (e) | PSNR: 28.41 dB -> 25.98 dB | Hurt |
| increase incident flux from 3 to 4 mean photons per frame (ablation) | naive estimator · 32×32 image, row (b) to row (e) | PSNR: 17.05 dB -> 16.44 dB | Hurt |
| increase modeled coincidence order through 2-, 3-, and 4-photon events (ablation) | multiphoton estimator · 32×32 reconstruction | MSE | Helped |
Surprising findings
- The naive estimator's PSNR drops only about 0.6 dB when mean photons per frame rises from 3 to 4, while the single-photon and multiphoton estimators drop about 3 dB each.
Limitations
- Models only up to four coincident photons; higher-order coincidences are ignored and introduce bias at high flux.
- Omits spatial priors, and the approximate estimator's bias is not analytically characterized.
- Demonstrated only with Monte Carlo simulation for one detector type and row-column architecture.
Experimental setup
Metrics: PSNR MSE Cramer-Rao bound comparison
Baselines: naive estimator single-photon estimator fully-addressed readout
Builds on: row-column readout multiplexing (Allman et al.) group testing surface codes (van den Berg et al.)
Closest papers by meaning
- bit2bit: 1-bit quanta video reconstruction via self-supervised photon prediction (NeurIPS 2024 · similarity 0.83)
- Resolution Limit of Single-Photon LiDAR (CVPR 2024 · similarity 0.82)
- Generalized Event Cameras (CVPR 2024 · similarity 0.82)
- Time-Efficient Light-Field Acquisition Using Coded Aperture and Events (CVPR 2024 · similarity 0.81)
- Towards 3D Vision with Low-Cost Single-Photon Cameras (CVPR 2024 · similarity 0.81)
- Separating the Wheat from the Chaff: Spatio-Temporal Transformer with View-interweaved Attention for Photon-Efficient Depth Sensing (AAAI 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.