Skip to content

relative-dn.md

Latest commit

 

History

History
88 lines (63 loc) · 4.97 KB

relative-dn.md

File metadata and controls

88 lines (63 loc) · 4.97 KB
Raw

Relative Characterization in DN

← Documentation index

This method describes how the response, the temporal noise and the spatial non-uniformity of a camera behave, from dark and flat frames. Results stay in digital numbers (DN): the method estimates no conversion gain, quantum efficiency or radiometric quantity. For the conversion gain, use the photon transfer curve.

Recipe: org.datalab.camera-characterization:relative-dn-characterization, version 1.1.0.

Run

Select the frames, then choose Plugins > Camera & Detector Characterization > Run camera characterization..., or use the Relative Camera characterization method of the Applications catalog. The quickstart is its demonstration.

Inputs

Slot Content
Dark frames At least 2 frames, shutter closed
Flat frames At least 4 frames of a uniform field, each with its exposure time
  • Each image holds one 2D frame. All frames have the same shape and data type, and finite values only.
  • Flat frames carry their exposure time, in seconds, in plugin.org.datalab.camera-characterization.exposure_time_s. Frames with the same exposure form one level. At least two levels must be unsaturated.
  • To set this key and the frame role on your own frames, see Use your own frames.

The inputs are checked before the computation. Errors stop the run. Warnings, such as zero or excessive temporal variance and excessive saturation, are kept with the results.

Parameters: minimum frame and level counts, saturation level and warning fraction, optional maximum temporal variance (0 disables it), aggregation block size, candidate-pixel threshold and histogram bins. These are workflow criteria, not acceptance limits.

Results

Output Type Content
response Signal Mean dark-subtracted signal versus exposure, with the metrics table
mean_dark Image Mean of all dark frames
mean_flat Image Mean of the highest level kept by the linear fit
dsnu_like_map Image Centered mean dark image, in DN
prnu_like_map Image Relative deviation of the dark-corrected flat image
candidate_pixel_map Image Pixels beyond the threshold in either map
prnu_row_profile, prnu_column_profile Signal Row and column means of the PRNU-like map
dsnu_distribution, prnu_distribution Signal Histograms of both maps

The metrics table gives the response slope and intercept, the dark temporal noise, the largest fit residual, the maximum unsaturated signal, the relative dynamic range, the selected flat exposure, the spatial non-uniformities, the candidate pixels and the saturation onset. Every status is Not assessed: no pass/fail standard is applied.

Method

For each level, the dark mean is subtracted from the flat mean. The temporal variance is the spatial mean of the per-pixel sample variance:

mean_signal_dn = mean(flat) - mean(dark)
temporal_variance_dn2 = mean_pixels(var_frames(pixel, ddof=1))
temporal_noise_dn = sqrt(temporal_variance_dn2)
relative_snr = mean_signal_dn / temporal_noise_dn
  • Response: a straight-line fit of mean_signal_dn versus exposure, on the levels below the saturation warning fraction. Residuals are given for every level.
  • Saturation onset: the first level that reaches the saturation warning fraction.
  • Dynamic range: maximum_unsaturated_signal_dn / dark_temporal_noise_dn.
  • Spatial maps: from the mean dark image D and the mean flat image F of the highest fitted level, dsnu_like_dn = D - mean(D) and prnu_like = (F - D) / mean(F - D) - 1. The non-uniformities are their standard deviations (ddof=1).
  • Candidate pixels: pixels beyond the sigma threshold in either map. This is a screening aid, not a defect classification.

A negative mean signal is kept, not clipped: it points to a problem in the data. A zero noise gives an infinite SNR or dynamic range.

Demonstration

The quickstart campaign shows every output. A separate synthetic campaign checks the method against the simulator truth: response slope within 0.5 %, dark noise within 2 % and fit residuals below 0.2 DN. A noiseless campaign checks the DSNU-like and PRNU-like maps against the truth maps. See Qualification.

Limits

  • Results are relative quantities in DN. The method does not claim EMVA 1288 compliance.
  • The PRNU-like map includes vignetting and dust: from frames alone, sensor PRNU and illumination cannot be separated.

Python

from datalab_camera_characterization.core import (
    CameraExposureSeries,
    CameraValidationParameters,
    characterize_relative_dn,
)

result = characterize_relative_dn(
    dark_frames,
    (
        CameraExposureSeries(flat_low, exposure_time_s=0.01),
        CameraExposureSeries(flat_high, exposure_time_s=0.02),
    ),
    CameraValidationParameters(saturation_dn=4095),
)