FiDeR: Denoising Astronomical Images at the Observational Resolution

arXiv:2611.01657 · astro-ph.IM · Submitted 2026-11-01 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "FiDeR: Denoising Astronomical Images at the Observational Resolution".

Jocelyn: Here's the gist of the paper. FiDeR presents Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF) We first apply a finite number of Richardson–Lucy (RL) iterations to the nonnegative source-plus-background image We then measure the response of a calibration point source processed with the same settings and derive a reconvolution kernel matched to the observational PSF We compare FiDeR with six denoisers using simulated point-source and mixed stellar–galaxy fields, and observations from the James Webb Space Telescope Near-Infrared Camera (NIRCam) and Mid-Infrared Instrument (MIRI) FiDeR introduces a method that calibrates the point-source response of the finite RL reconstruction at the relevant background-to-source ratio and uses that response to derive the reconvolution kernel The target is normally the input PSF The method reconstructs source and background together, then subtract the background after reconvolution Finite-Iteration Reconstruction The reconstruction involves a Poisson negative log likelihood L(x; y) and the RL update x(j+1) = x(j)⊙ HT h y Hx(j)i, with x(0) = y The tests use iteration budgets of 200 or 1000 Point-Source Response and Matching Kernel The normalized background-subtracted response is defined as K(β)e,k = Dk(Fch + b1; h) − b1 Fc = Dk(h + β1; h) − β, where β = b/Fc A regularized Fourier solution is used to seek a kernel ak such that ak ∗ K(β)e,k ≃ h⋆, where the target h⋆ is normally the input PSF h Simulation Results and Comparison FiDeR reaches peak signal-to-noise ratios of 46.701 and 44.850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.657 and 0.858 counts In the mixed field, FiDeR reaches peak signal-to-noise ratios of 46.701 and 44.850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.657 and 0.858 counts FiDeR’s aperture-flux MAEs are 0.96% and 1.06%, compared with 1.04% and 1.18% for BM3D Observational Validation The NIRCam observations show lower core and combined-region errors The threeexposure MIRI test gives benefits mainly in the stellar aperture and photometry FiDeR reduces core RMSE to 16.598 counts, from 35.366 for Raw and 35.304 for BM3D Within the stellar aperture, FiDeR gives the lowest RMSE, 0.1230 ± 0.0157 MJy sr−1 FiDeR achieves substantial noise reduction with little change in the measured PSF in the tested regimes, and improves on conventional denoisers in several scenes These properties can benefit repeated exposures before UPDC deconvolution coaddition (L.

Vera: Wang et al.

Jocelyn: 2022, 2025), machine-learning inputs, and morphological measurements The expected gains in these applications should be measured with end-to-end tests that account for the covariance of the processed pixels APPENDIX A.

Vera: DIAGNOSTIC DEFINITIONS For an output source image I and reference T, we use the residual di = Ii − Ti We use 11 × 11 Gaussian windows with standard deviation 1.5 pixels and population covariance The structural similarity index measure (SSIM; Z.

Paper summary: Vera: So to recap for this paper, FiDeR is a new method designed to reduce noise in astronomical images while keeping approximately the input point spread function.

Jocelyn: The main claim is that they first apply a finite number of Richardson–Lucy iterations to an image that contains both a source and some background.

Subrahmanyan: Then they measure the response of a calibration point source processed with those same settings, and use that measurement to derive a reconvolution kernel matched to the observational point spread function.

Vera: The whole thesis is that this approach lets them reconstruct the source and background together, and then they subtract just the background after reconvolution.

Jocelyn: They compare FiDeR with six other denoisers using simulated fields involving both point sources and mixed stellar and galaxy fields to see how it stacks up against them.

Subrahmanyan: Furthermore, they apply this exact approach to James Webb Space Telescope observations.

Vera: The results show that FiDeR achieves substantial noise reduction with very little change in the measured point spread function in the tested regimes.

Jocelyn: This means it has potential benefits for repeated exposures before deconvolution coaddition, machine learning inputs, and morphological measurements.

Subrahmanyan: The method extends previous uses of PSF-based iterative processing to a new task: denoising at the observational resolution instead of just upsampling or deconvolution coaddition.

Conclusion: Vera: So, looking at "FiDeR: Denoising Astronomical Images at the Observational Resolution" by Wang and colleagues, it seems this paper presents a way to clean up noisy astronomical data while trying to preserve the original image structure.

Jocelyn: What they’re really saying is that this technique calibrates how a finite Richardson–Lucy reconstruction responds to the specific background and source ratio we're dealing with.

Subrahmanyan: By using that calibration, they build a kernel that helps them separate the background from the source after reconstruction.

Vera: The implication is that for people working with repeated exposures or when using machine learning on astronomical data, this could give them cleaner inputs for analysis.

Jocelyn: It’s about making sure the noise reduction doesn't come at a huge cost to the actual shape of the object we're studying.

Subrahmanyan: The paper suggests that because it works across both compact and extended sources in their tests, this is a technique that could be useful in many different types of imaging science.

Lei Wang, Bin Li, Lin Nie, Zhaojun Yan, Xuechun Chen, Yachong Diao, Huanyuan Shan

Purple Mountain Observatory, Chinese Academy of Sciences · Shanghai Astronomical Observatory, Chinese Academy of Sciences · Beijing City University

astro-ph.IM

Submitted: 2026-11-01

Updated: 2026-11-01

Comments: 13 pages, 5 figures, 4 tables

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: FiDeR presents Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF)…

Key concepts

Finite-Iteration Reconstruction
This involves using a finite number of Richardson-Lucy iterations to reconstruct an image from a Poisson negative log likelihood. This process iteratively refines an initial guess until it converges, with test budgets varying between 200 and 1000 iterations.
Point-Source Response and Matching Kernel
FiDeR measures how the reconstructed point source responds at different background-to-source ratios. This response is used to derive a reconvolution kernel that is matched to the observational Point Spread Function (PSF), aiming to reconstruct the true input PSF.
Reconstruction Process
The method first reconstructs both the source and background components simultaneously using the finite RL update rule. After this joint reconstruction, the background is subtracted from the result via reconvolution, which helps separate noise from actual signal structures.
PSF Preservation
The primary goal of FiDeR is to reduce noise while maintaining approximately the original input point-spread function (PSF). This means that after denoising, the characteristic shape and resolution of the astronomical source remain largely unchanged.

Terminology

Summary

FiDeR presents Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF) We first apply a finite number of Richardson–Lucy (RL) iterations to the nonnegative source-plus-background image We then measure the response of a calibration point source processed with the same settings and derive a reconvolution kernel matched to the observational PSF We compare FiDeR with six denoisers using simulated point-source and mixed stellar–galaxy fields, and observations from the James Webb Space Telescope Near-Infrared Camera (NIRCam) and Mid-Infrared Instrument (MIRI)

Methodology of FiDeR

FiDeR introduces a method that calibrates the point-source response of the finite RL reconstruction at the relevant background-to-source ratio and uses that response to derive the reconvolution kernel The target is normally the input PSF The method reconstructs source and background together, then subtract the background after reconvolution

Finite-Iteration Reconstruction

The reconstruction involves a Poisson negative log likelihood L(x; y) and the RL update x(j+1) = x(j)⊙ HT h y Hx(j)i, with x(0) = y The tests use iteration budgets of 200 or 1000

Point-Source Response and Matching Kernel

The normalized background-subtracted response is defined as K(β)e,k = Dk(Fch + b1; h) − b1 Fc = Dk(h + β1; h) − β, where β = b/Fc A regularized Fourier solution is used to seek a kernel ak such that ak ∗ K(β)e,k ≃ h⋆, where the target h⋆ is normally the input PSF h

Simulation Results and Comparison

FiDeR reaches peak signal-to-noise ratios of 46.701 and 44. p. 0

Where the paper says it

For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46. 701 and 44.

p. 0
Where the paper says it

For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46. 701 and 44.

p. 0
Where the paper says it

For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46. 701 and 44.

p. 0
Where the paper says it

For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46. 701 and 44.

850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0. p. 0
Where the paper says it

701 and 44. 850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.

p. 0
Where the paper says it

701 and 44. 850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.

657 and 0.858 counts In the mixed field, FiDeR reaches peak signal-to-noise ratios of 46. p. 0
Where the paper says it

In the stellar tests, brightsource widths remain close to the input widths; faint sources show small residual offsets. For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46.

701 and 44.850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.657 and 0.858 counts FiDeR’s aperture-flux MAEs are 0. p. 5
Where the paper says it

580 FiDeR 46. 463 FiDeR’s aperture-flux MAEs are 0.

96% and 1.06%, compared with 1. p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

p. 4
Where the paper says it

Wing RMSE falls to 0. 858 counts, compared with BM3D’s 1.

04% and 1.18% for BM3D

Observational Validation

The NIRCam observations show lower core and combined-region errors The threeexposure MIRI test gives benefits mainly in the stellar aperture and photometry FiDeR reduces core RMSE to 16.598 counts, from 35. p. 7

Where the paper says it

FiDeR reduces core RMSE to 16. 598 counts, from 35.

p. 7
Where the paper says it

FiDeR reduces core RMSE to 16. 598 counts, from 35.

366 for Raw and 35. p. 7
Where the paper says it

598 counts, from 35. 366 for Raw and 35.

p. 7
Where the paper says it

598 counts, from 35. 366 for Raw and 35.

304 for BM3D Within the stellar aperture, FiDeR gives the lowest RMSE, 0. p. 7
Where the paper says it

Thus FiDeR improves the raw image but does not lead the full-field comparison. Within the stellar aperture, FiDeR gives the lowest RMSE, 0.

1230 ± 0.0157 MJy sr−1

Conclusion and Applications

FiDeR achieves substantial noise reduction with little change in the measured PSF in the tested regimes, and improves on conventional denoisers in several scenes These properties can benefit repeated exposures before UPDC deconvolution coaddition (L. Wang et al. p. 0

Where the paper says it

arXiv preprint Typeset using LATEX default style in AASTeX7. 2 FiDeR: Denoising Astronomical Images at the Observational Resolution Lei Wang, 1 Bin Li,1 Lin Nie,2 Zhaojun Yan,2 Xuechun Chen,3 Yachong Diao,3 and Huanyuan Shan2 1Purple Mountain Observatory, Chinese Academy of Sciences, 10 Yuanhua Road, Qixia District, Nanjing 210023, China 2Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, People’s Republic of China 3Beijing City University, No.

p. 0
Where the paper says it

arXiv preprint Typeset using LATEX default style in AASTeX7. 2 FiDeR: Denoising Astronomical Images at the Observational Resolution Lei Wang, 1 Bin Li,1 Lin Nie,2 Zhaojun Yan,2 Xuechun Chen,3 Yachong Diao,3 and Huanyuan Shan2 1Purple Mountain Observatory, Chinese Academy of Sciences, 10 Yuanhua Road, Qixia District, Nanjing 210023, China 2Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, People’s Republic of China 3Beijing City University, No.

2022, 2025), machine-learning inputs, and morphological measurements The expected gains in these applications should be measured with end-to-end tests that account for the covariance of the processed pixels

APPENDIX A. DIAGNOSTIC DEFINITIONS

For an output source image I and reference T, we use the residual di = Ii − Ti We use 11 × 11 Gaussian windows with standard deviation 1.5 pixels and population covariance The structural similarity index measure (SSIM; Z. p. 2

Where the paper says it

These comparisons use the implementations and settings specified in Appendix D. We measure image root mean square error (RMSE), peak signal-to-noise ratio (PSNR), structural similarity (SSIM; Z.

Wang et al. 2004) is SSIM(I, T) = (2µIµT + C1)(2σIT + C2)((µ2I + µ2T + C1)(σ2I + σ2T + C2) The simulated flux mean absolute error (MAE) is MAEF = 100/Nsrc X s(Fs(I) − Fs(T))/Fs(T) For stellar widths, we fit a circular Gaussian plus a constant to both output and reference, with a fixed true center and r ≤ 8 pixels

B. CONSERVATION AND CALIBRATED RESPONSE

For positive predicted intensities and normalized convolution on a complete periodic or otherwise flux-complete domain, X i x(j+1) i = X i x(j) i / (X l Hli yl (Hx(j))l)(B7)=X l yl. Applying this identity to the calibration image and subtracting b gives PK(β)e,k = 1

C. BACKGROUND AND NOISE TREATMENT

We use Ec = cE, beff = c squared vbg, yeff = Ec + beff1 We apply Equation 6, and convert back by 1/c

D. IMPLEMENTATION AND REFERENCE CONVENTIONS

The spatial filter labeled Kalman forms the mean mi and variance vi of the eight neighbors, clips negative roundoff variance to zero, and returns mi + [vi/(vi + σ2 n)](Ei − mi) The implementation uses BM3D 4.0.3 and its BM4D dependency 4.2.5 For an aperture split, the combined squared RMSE obeys RMSE squared c+w = NcRMSE squared c + NwRMSE squared w/(Nc + Nw) The raw-image identity illustrates the correlation introduced by a self-including reference

REFERENCES

Anderson, J., & King, I. p. 1

Where the paper says it

Anderson & I. King 2000).

R. 2000, PASP, 112, 1360, doi: 10. p. 1
Where the paper says it

Anderson & I. King 2000).

1086/316632 Aniano, G. p. 0
Where the paper says it

Magain et al. Aniano et al.

, Draine, B. T., Gordon, K. D., & Sandstrom, K. 2011, PASP, 123, 1218, doi: 10. p. 0
Where the paper says it

Buades et al. 2005, 2011; K.

1086/662219 Buades, A. p. 0
Where the paper says it

Non-local means (NLM) and blockmatching and three-dimensional filtering (BM3D) use similarities between image patches to suppress noise (A. Buades et al.

, Coll, B., & Morel, J.-M. 2005, Multiscale Modeling & Simulation, 4, 490, doi: 10. p. 0
Where the paper says it

Buades et al. 2005, 2011; K.

1137/040616024 Buades, A. p. 0
Where the paper says it

Non-local means (NLM) and blockmatching and three-dimensional filtering (BM3D) use similarities between image patches to suppress noise (A. Buades et al.

, Coll, B., & Morel, J.-M. 2011, Image Processing On Line, 1, 208, doi: 10. p. 0
Where the paper says it

The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations. Keywords: Astronomical image processing — Deconvolution — Point spread function — Photometry — Astronomy data analysis 1.

5201/ipol.2011.bcm nlm Dabov, K. p. 0
Where the paper says it

2005, 2011; K. Dabov et al.

, Foi, A., Katkovnik, V., & Egiazarian, K. 2007, IEEE Transactions on Image Processing, 16, 2080, doi: 10. p. 0
Where the paper says it

The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations. Keywords: Astronomical image processing — Deconvolution — Point spread function — Photometry — Astronomy data analysis 1.

1109/TIP.2007.901238 Donoho, D. p. 0
Where the paper says it

Lee 1980; D. Donoho 1995).

L. 1995, IEEE Transactions on Information Theory, 41, 613, doi: 10. p. 0
Where the paper says it

Lee 1980; D. Donoho 1995).

1109/TIP.4766994 Fruchter, A. p. 5
Where the paper says it

We register the exposures on a common grid, use compatible intensity units and background subtraction, and construct Drizzle-based coadds (A. Fruchter & R.

S., & Hook, R. p. 5
Where the paper says it

Fruchter & R. Hook 2002).

N. 2002, PASP, 114, 144, doi: 10. p. 5
Where the paper says it

Fruchter & R. Hook 2002).

1086/338393 Kawata, S. p. 0
Where the paper says it

Restoration followed by reblurring, controlled-resolution deconvolution, and common-resolution kernels are established approaches (S. Kawata & Y.

, & Ichioka, Y. p. 0
Where the paper says it

Kawata & Y. Ichioka 1980; P.

1980, Journal of the Optical Society of America, 70, 768, doi: 10. p. 0
Where the paper says it

Adaptive filters and wavelet methods adjust their response to local structure or scale (J. Lee 1980; D.

1364/JOSA.70.000768 Lee, J. p. 0
Where the paper says it

Adaptive filters and wavelet methods adjust their response to local structure or scale (J. Lee 1980; D.

-S. 1980, IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-2, 165, doi: 10. p. 0
Where the paper says it

The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations. Keywords: Astronomical image processing — Deconvolution — Point spread function — Photometry — Astronomy data analysis 1.

1109/TPAMI.4766994 Lucy, L. p. 0
Where the paper says it

26 Muyan Road (Yangzhen Section), Yangzhen, Shunyi District, Beijing 101309, People’s Republic of China ABSTRACT We present Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF). We first apply a finite number of Richardson–Lucy (RL) iterations to the nonnegative source-plusbackground image.

B. 1974, AJ, 79, 745, doi: 10. p. 0
Where the paper says it

Richardson 1972; L. Lucy 1974).

1086/338393 Magain, P. p. 0
Where the paper says it

Ichioka 1980; P. Magain et al.

, Courbin, F., & Sohy, S. 1998, ApJ, 494, 472, doi: 10.1086/305187 M¨akinen, Y., Azzari, L., & Foi, A. 2020, IEEE Transactions on Image Processing, 29, 8339, doi: 10. p. 0
Where the paper says it

The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations. Keywords: Astronomical image processing — Deconvolution — Point spread function — Photometry — Astronomy data analysis 1.

1109/TIP.2020.3014721 M¨akitalo, M., & Foi, A. 2011, IEEE Transactions on Image Processing, 20, 99, doi: 10. p. 0
Where the paper says it

The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations. Keywords: Astronomical image processing — Deconvolution — Point spread function — Photometry — Astronomy data analysis 1.

1109/TIP.2056693 Prato, M. p. 0
Where the paper says it

Stopping the iteration before convergence provides regularization (M. Prato et al.

, Cavicchiolioli R., Zanni L., Boccacci P., & Bertero M. 2012, A&A, 539, A133, doi: 10. p. 0
Where the paper says it

Prato et al. 2012), but also leaves a finite-iteration point-source response.

1051/0004-6361/2011868 Richardson, W. p. 0
Where the paper says it

26 Muyan Road (Yangzhen Section), Yangzhen, Shunyi District, Beijing 101309, People’s Republic of China ABSTRACT We present Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF). We first apply a finite number of Richardson–Lucy (RL) iterations to the nonnegative source-plusbackground image.

H. 1972, Journal of the Optical Society of America, 62, 55, doi: 10. p. 0
Where the paper says it

Richardson–Lucy (RL) reconstruction uses a known PSF and a Poisson likelihood to recover a nonnegative image (W. Richardson 1972; L.

1364/JOSA.62.000055 Wang, L. p. 0
Where the paper says it

arXiv preprint Typeset using LATEX default style in AASTeX7. 2 FiDeR: Denoising Astronomical Images at the Observational Resolution Lei Wang, 1 Bin Li,1 Lin Nie,2 Zhaojun Yan,2 Xuechun Chen,3 Yachong Diao,3 and Huanyuan Shan2 1Purple Mountain Observatory, Chinese Academy of Sciences, 10 Yuanhua Road, Qixia District, Nanjing 210023, China 2Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, People’s Republic of China 3Beijing City University, No.

, Li, G., & Kang, X. 2022, MNRAS, 517, 787, doi: 10. p. 1
Where the paper says it

Wang et al. 2022), UPdec-Webb (L.

1093/mnras/stac2664 Wang, L. p. 0
Where the paper says it

arXiv preprint Typeset using LATEX default style in AASTeX7. 2 FiDeR: Denoising Astronomical Images at the Observational Resolution Lei Wang, 1 Bin Li,1 Lin Nie,2 Zhaojun Yan,2 Xuechun Chen,3 Yachong Diao,3 and Huanyuan Shan2 1Purple Mountain Observatory, Chinese Academy of Sciences, 10 Yuanhua Road, Qixia District, Nanjing 210023, China 2Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, People’s Republic of China 3Beijing City University, No.

, Shan, H., Nie, L., et al. 2025, ApJS, 276, 36 Zhang, S. p. 1
Where the paper says it

2025), and filtered image deconvolution and noise backfilling (FiDNoBa; S. Zhang et al.

, Wang, L., Diao, Y., et al. 2026, AJ, 172, 67

--- Page 4 ---

Mock-I: 36 point sources

The crowded field contains a 6 × 6 grid on a 128 × 128 image, with 20-pixel separations and source fluxes of 6000–11,280 counts FiDeR suppresses the visible pixel fluctuations; its residuals are mainly coherent, low-amplitude variations rather than high-frequency noise FiDeR reduces pixel fluctuations but leaves low-amplitude spatial structure

Mock Mock-I Analysis

FiDeR suppresses the visible pixel fluctuations; its residuals are mainly coherent, low-amplitude variations rather than high-frequency noise FiDeR reduces pixel fluctuations but leaves low-amplitude spatial structure

Mock Mock-II Analysis

FiDeR suppresses pixel-scale fluctuations while retaining the main source geometries FiDeR reduces fluctuations around both compact and extended sources in Figure 3

Performance Metrics

FiDeR gives PSNRs of 46.701 and 44.850 dB at b = 10 and 20, exceeding BM3D by 10. p. 4

Where the paper says it

701 and 44. 850 dB at b = 10 and 20, exceeding BM3D by 10.

p. 4
Where the paper says it

701 and 44. 850 dB at b = 10 and 20, exceeding BM3D by 10.

511 and 8.301 dB FiDeR’s aperture-flux MAEs are 0. p. 5
Where the paper says it

580 FiDeR 46. 463 FiDeR’s aperture-flux MAEs are 0.

p. 5
Where the paper says it

580 FiDeR 46. 463 FiDeR’s aperture-flux MAEs are 0.

96% and 1.06%, compared with 1.04% and 1.18% for BM3D FiDeR reduces fluctuations around both compact and extended sources in Figure 3

JWST Observations

FiDeR reduces core RMSE to 16.598 counts, from 35.366 for Raw and 35.304 for BM3D FiDeR has the highest combined SSIM, 0. p. 7

Where the paper says it

781 counts overall, consistent with its loss of weak structure in this cutout. FiDeR has the highest combined SSIM, 0.

9998485, compared with 0. p. 7
Where the paper says it

FiDeR has the highest combined SSIM, 0. 9998485, compared with 0.

9997806 for BM3D

MIRI Observations

FiDeR reduces full-image RMSE from Raw’s 0.1461 to 0.1116 MJy sr−1 and raises PSNR from 59. p. 7

Where the paper says it

FiDeR reduces full-image RMSE from Raw’s 0. 1116 MJy sr−1 and raises PSNR from 59.

991 to 62.215 dB Within the stellar aperture, FiDeR gives the lowest RMSE, 0. p. 7
Where the paper says it

Thus FiDeR improves the raw image but does not lead the full-field comparison. Within the stellar aperture, FiDeR gives the lowest RMSE, 0.

1230 ± 0.0157 MJy sr−1

Table 3 Summary

FiDeR gives PSNRs of 46.701 and 44.850 dB at b = 10 and 20, exceeding BM3D by 10.511 and 8.301 dB FiDeR’s aperture-flux MAEs are 0.96% and 1.06%, compared with 1.04% and 1.18% for BM3D FiDeR reduces fluctuations around both compact and extended sources in Figure 3

Table 4 Summary

FiDeR’s aperture-flux MAEs are 0.96% and 1.06%, compared with 1.04% and 1.18% for BM3D FiDeR reduces fluctuations around both compact and extended sources in Figure 3

Table 4 Summary (Observation-II)

FiDeR’s aperture-flux MAEs are 0.96% and 1.06%, compared with 1.04% and 1.18% for BM3D FiDeR reduces fluctuations around both compact and extended sources in Figure 3

FiDeR’s aperture-flux MAEs are 0.96% and 1.06%, compared with 1.04% and 1.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved noise reduction in astronomical images by applying Finite-iteration Deconvolution–Reconvolution Denoising (FiDeR) to reduce noise while retaining approximately the input point-spread function (PSF).

  2. Ability to perform denoising at observational resolution, as FiDeR is tested for this specific task, and can provide inputs for further reconstruction and measurement.

  3. Enhanced performance in mixed stellar–galaxy fields by reaching peak signal-to-noise ratios of 46.701 and 44.850 dB at backgrounds of 10 and 20 counts per pixel, outperforming six other denoisers in key diagnostics like wing errors, aperture fluxes, and shape errors.

  4. Improved fidelity in stellar profile analysis by showing that bright source widths remain close to the input widths with scatters of 0.02–0.04 percentage points, indicating good preservation of source morphology for bright sources.

  5. More robust handling of faint sources, as FiDeR shows that faint sources show small residual offsets, suggesting it retains better structural information for lower signal-to-noise data compared to some methods like the mean filter which broadens every tested PSF.

  6. Enhanced image quality metrics in NIRCam observations, as FiDeR reduces core and combined-region errors compared to other methods, with a combined RMSE of 7.411 counts versus BM3D's 13.408 counts in the full-field comparison.

  7. Superior performance for stellar aperture photometry and morphology in MIRI data, as FiDeR gives the lowest RMSE [MJy/sr], 0.1230 ± 0.0157 MJy sr−1, within the stellar aperture compared to BM3D's 0.1292 ± 0.0105 MJy sr−1 and a higher PSNR of 61.405 ± 1.077 dB against Raw exposure noise.

  8. Better preservation of source characteristics in MIRI data, as FiDeR shows stellar aperture flux changes by 0.005 ± 0.013% relative to the same raw exposure, which is superior to the changes reported by Median and NLM filtering methods for galaxy apertures in that regime.

  9. Capability to support subsequent processing steps, such as inputs for machine-learning or UPDC deconvolution coaddition, as FiDeR's residual can provide the filtered image, with its corresponding residual used for noise backfilling.

Abstract

We present Finite-iteration Deconvolution--Reconvolution Denoising (FiDeR), a method for reducing noise in astronomical images while retaining approximately the input point-spread function (PSF). We first apply a finite number of Richardson--Lucy (RL) iterations to the nonnegative source-plus-background image. We then measure the response of a calibration point source processed with the same settings and derive a reconvolution kernel matched to the observational PSF. The calibration includes the background-to-source flux ratio. We compare FiDeR with six denoisers using simulated point-source and mixed stellar--galaxy fields, and observations from the James Webb Space Telescope Near-Infrared Camera (NIRCam) and Mid-Infrared Instrument (MIRI). In the stellar tests, bright-source widths remain close to the input widths; faint sources show small residual offsets. For the mixed field, FiDeR reaches peak signal-to-noise ratios of 46.701 and 44.850 dB at backgrounds of 10 and 20 counts per pixel, with wing errors of 0.657 and 0.858 counts. It also gives small aperture-flux and shape errors, although block-matching and three-dimensional filtering (BM3D) gives higher structural similarity. The NIRCam data show lower core and combined-region errors, while the three-exposure MIRI test gives benefits mainly in the stellar aperture and photometry. These results support substantial noise reduction with little change in the measured PSF within the tested regimes. The denoised images can provide inputs for further reconstruction and measurement; their uncertainties must account for residual noise correlations.

Sources

Related papers