Photon Transfer Curve Calculation

Sensor characterization

Estimate system gain, read noise, full well capacity, dynamic range, and noise behavior from flat and bias frame statistics using a clean Janesick style photon transfer workflow.

PTC Calculator

Enter paired flat field and bias measurements. This calculator uses the standard pair-difference method where variance is derived from the difference image and the mean signal is bias-subtracted.

Method used by this calculator

1. Average the two flat means and the two bias means.
2. Subtract bias mean from flat mean to get usable signal in DN.
3. Convert flat pair difference standard deviation to variance with variance = sigma² / 2.
4. Estimate gain with gain = signal / variance, in electrons per DN.
5. Estimate read noise from the bias pair using sigma_bias / square root of 2.
6. Compute full well, dynamic range, and noise curves in electrons.

Results

Calculated sensor characteristics from your PTC inputs.

Expert Guide to Photon Transfer Curve Calculation

Photon transfer curve calculation is one of the most useful techniques in scientific imaging because it links the digital output of a sensor to the underlying physics of photoelectron generation. If you work with CCDs, CMOS detectors, astronomy cameras, microscopy systems, machine vision sensors, or calibrated laboratory imagers, the photon transfer curve, often abbreviated PTC, gives you a compact but powerful way to measure gain, read noise, shot noise behavior, saturation response, and dynamic range. In practical terms, it tells you how efficiently the detector converts photons into measurable signal and how noise grows as illumination increases.

The core idea is straightforward. Light arrives at the detector and creates photoelectrons. Those electrons are then converted by the camera electronics into digital numbers, or DN. A photon transfer calculation studies the relationship between signal mean and signal variance across a range of exposures. At low signal, read noise often dominates. At medium signal, shot noise becomes dominant and variance scales approximately linearly with signal. Near saturation, the response deviates because the pixel wells approach capacity and the detector can no longer store additional charge linearly.

This makes PTC analysis far more than an academic exercise. It is a practical calibration method. It helps imaging scientists answer questions such as these: What is the system gain in electrons per DN? How much read noise does the detector contribute? What is the full well capacity? What dynamic range is available before saturation? Does the camera behave linearly enough for photometry or metrology? When you calculate the photon transfer curve correctly, you can optimize exposure settings, compare sensors objectively, and validate whether the camera meets performance specifications.

What a Photon Transfer Curve Measures

A complete photon transfer curve usually plots noise or variance against mean signal. The most common interpretations are:

• Gain: the conversion factor between digital counts and electrons. If gain is 1.2 e-/DN, then each digital count corresponds to about 1.2 electrons.
• Read noise: the electronic noise floor introduced during charge conversion and readout. This appears most strongly when illumination is very low.
• Shot noise: the unavoidable statistical variation associated with photon arrival. For a signal of N electrons, the standard deviation is approximately the square root of N.
• Full well capacity: the maximum number of electrons a pixel can store before strong nonlinearity or saturation occurs.
• Dynamic range: the ratio of maximum useful signal to the read noise floor.

Because a PTC ties all of these together, it is widely used in astronomy, remote sensing, semiconductor inspection, biological imaging, and detector manufacturing. The same physics applies whether the device is observing stars, cells, wafers, or calibration targets.

Why Paired Images Are Used

A reliable photon transfer curve calculation generally uses paired images rather than a single frame. Two flat frames are captured at the same illumination level, and two bias or dark frames are captured for the noise floor. The reason is statistical isolation. If you subtract one flat from another, fixed pattern structure is largely removed, while random noise remains. The variance of the difference image is then related to the variance of one frame by a factor of two. This is why the calculator above uses variance = sigma² / 2 when sigma is the standard deviation of the difference image.

This pair-difference method is standard because it reduces contamination from pixel response nonuniformity. Without differencing, a textured or imperfectly illuminated flat field can inflate the measured variance and lead to a gain estimate that is too low. The same logic applies to the bias pair. Subtracting two bias frames removes fixed offsets and leaves random read noise, from which single-frame read noise is estimated as sigma_bias_difference divided by the square root of 2.

The Fundamental Equations

For a simple Janesick style calculation at one flat field level, the essential equations are:

1. Average flat signal: mean_flat = (flat1 + flat2) / 2
2. Average bias level: mean_bias = (bias1 + bias2) / 2
3. Bias-subtracted signal: signal_DN = mean_flat – mean_bias
4. Flat variance from difference image: variance_DN = sigma_flat_difference² / 2
5. System gain: gain = signal_DN / variance_DN, in e-/DN
6. Read noise in DN: read_noise_DN = sigma_bias_difference / square root of 2
7. Read noise in electrons: read_noise_e = gain × read_noise_DN

Once gain is known, many other metrics follow immediately. Full well can be estimated from the saturation level in DN multiplied by the gain. Dynamic range is often reported as full_well_e divided by read_noise_e, while the dynamic range in dB is 20 × log10 of that ratio.

Important practical note: a single-point estimate is useful, but a full photon transfer characterization is best done across many illumination levels. That produces the full curve, reveals linearity limits, and helps separate shot-noise-dominated behavior from low-signal electronics and high-signal saturation effects.

How to Interpret the Output

If your calculated gain is close to the camera’s specification, that is a good sign that your data collection and variance extraction are sound. If the read noise is much higher than expected, several issues may be present: poor grounding, higher readout speed, temperature instability, pattern noise leaking into the measurement, compression in the camera path, or incorrect use of dark instead of true bias frames. If the dynamic range looks unrealistically large, check whether the saturation DN is set too high or whether the read noise estimate was suppressed by overprocessing.

A healthy photon transfer curve usually has three recognizable regions. First, at very low signal, the curve is flattened by read noise because electronics dominate. Second, at moderate signal, the relationship between variance and mean becomes nearly linear, indicating shot-noise-limited performance. Third, as the sensor approaches capacity, linearity breaks down and the curve bends because pixels can no longer store charge proportionally.

Comparison Table: Digital Depth and ADC Range

One factor that often confuses new users is the relationship between bit depth and digital saturation. The table below lists exact maximum DN values for common ADC depths. These are not sensor-specific estimates; they are exact digital limits implied by the converter width.

ADC Bit Depth	Maximum DN Value	Total Digital Levels	Typical Use Case
12-bit	4,095	4,096	High speed machine vision and embedded imaging
14-bit	16,383	16,384	Many astronomy and microscopy cameras
16-bit	65,535	65,536	Scientific CCD and sCMOS acquisition pipelines
18-bit	262,143	262,144	Precision instrumentation and digitizers

These digital limits matter because a PTC result expressed in electrons depends on both gain and the highest usable DN before nonlinearity. If the analog front end saturates well before the ADC maximum, then the true full well in electrons may be lower than a simple max-DN conversion implies. In other words, the ADC limit is only part of the story. The sensor itself may clip first.

Representative Detector Performance Statistics

The values below are representative published figures commonly cited in major astronomy detector documentation. Exact values vary by operating mode, gain setting, readout speed, detector temperature, and calibration release, but the table gives useful scale for what real scientific imagers look like in practice.

Detector System	Reported Read Noise	Reported Full Well	Context
Hubble WFC3 UVIS CCD	About 3.1 e-	About 70,000 e-	Space-based scientific CCD performance from STScI documentation
Hubble ACS WFC CCD	About 4 to 5 e-	About 84,000 e-	Representative CCD imaging performance from STScI instrument references
JWST NIRCam detector chain	Roughly 10 to 15 e- depending on mode	About 65,000 e- scale capacity	Infrared array performance reported in observatory documentation

These numbers are valuable because they show how meaningful a few electrons of read noise can be. In ultra-low-light work, reducing read noise from 5 e- to 2 e- can substantially improve detectability. In higher-flux applications, full well and linearity may matter more because they determine how much dynamic range the sensor can deliver before the response compresses.

Step by Step Workflow for Accurate Photon Transfer Curve Calculation

1. Stabilize the system. Allow the camera temperature, electronics, and illumination source to reach a stable operating condition.
2. Capture paired bias frames. Use zero or minimal exposure with shutter closed if appropriate to measure baseline read behavior.
3. Capture paired flat frames. Keep illumination spatially uniform and exposure conditions identical for each pair.
4. Compute means. Measure the average DN of each flat and bias frame over a clean region of interest, avoiding dust shadows and defective pixels.
5. Create difference images. Subtract flat1 from flat2 and bias1 from bias2, then compute their standard deviations.
6. Calculate gain and read noise. Use the equations shown above.
7. Repeat at multiple signal levels. Build the full curve to verify linearity and identify the shot-noise-dominated region.
8. Estimate full well and dynamic range. Use the highest linear or saturating signal along with the noise floor.

Common Mistakes That Distort PTC Results

• Illumination flicker: unstable lamps or displays introduce extra variance not caused by the sensor.
• Insufficient flat uniformity: gradients and patterns can mimic signal-dependent noise if difference processing is not done correctly.
• Compression or camera processing: gamma, denoising, and hidden image processing invalidate photon transfer assumptions.
• Mixing operating modes: changing gain mode, readout rate, or binning between exposures breaks comparability.
• Incorrect region of interest: hot pixels, cosmic ray events, amplifier boundaries, and vignetted edges can bias mean and variance.
• Too few data points: a single level can estimate gain, but it cannot reveal where linearity fails.

Why Photon Transfer Curves Matter in Real Applications

In astronomy, photon transfer calculations support detector calibration for photometry, astrometry, and faint-object detection. In fluorescence microscopy, they help determine whether the imaging chain is limited by shot noise or by electronics, which directly affects exposure planning and dose minimization. In machine vision and industrial metrology, a well-characterized PTC helps engineers predict measurement uncertainty, optimize lighting, and compare sensors across vendors using physics-based metrics instead of marketing labels.

Another important use case is system validation after integration. A camera might meet vendor specifications on a lab bench, yet show degraded performance when installed in a larger instrument because of EMI, thermal drift, firmware settings, or interface bandwidth limits. A fresh photon transfer calculation can identify these issues quickly. If read noise rises or gain shifts unexpectedly, the calibration provides immediate evidence that the system behavior changed.

How to Use This Calculator Effectively

The calculator on this page is designed for a practical single-level estimate using paired flats and paired bias frames. That makes it ideal for quick checks, commissioning work, and educational demonstrations. For production characterization, repeat the same process at many signal levels and plot the measured points. The linear slope region is where gain extraction is most reliable. Very low signal can be dominated by read noise and quantization effects. Very high signal can be affected by nonlinearity, blooming, clipping, and well saturation.

When you compare multiple cameras, always match operating conditions: temperature, readout rate, analog gain mode, ADC depth, exposure timing, and processing path. A fair photon transfer comparison requires the same test discipline for every device.

Authoritative References

For deeper reading and published detector performance context, review these authoritative sources:

• Space Telescope Science Institute: WFC3 Instrument Handbook
• Space Telescope Science Institute: ACS Data Handbook
• NIST Detector Calibration Resources

Final Takeaway

Photon transfer curve calculation is the bridge between what a detector records and what the incoming light actually did at the pixel. By measuring mean signal, variance, and noise floor carefully, you can derive gain, read noise, full well, and dynamic range in a way that is grounded in detector physics rather than guesswork. That is why the method remains one of the most trusted calibration tools in scientific imaging. If you capture clean paired data and interpret the resulting curve with care, the PTC becomes a precise map of sensor behavior from the noise floor to saturation.

Note: published performance figures in major observatory documentation can vary by amplifier, readout mode, calibration epoch, and operating temperature. Use your instrument’s current handbook or engineering data for formal acceptance testing.