Gamma and transfer functions. Numbers have a direction.

A linear signal is proportional to a defined light quantity: moving from 0.10 to 0.20 doubles that quantity while its normalisation remains fixed. It does not mean twice the perceived lightness or necessarily 10 and 20 cd/m². Establish whether the values refer to scene light, relative output or absolute display luminance.

Adding light and ordinary averaging for a blur have a physical interpretation in linear light. Linearity alone does not locate an image relative to output rendering. An image already prepared for a display can also be represented linearly.

Integer storage has a finite number of levels. Nonlinear encoding redistributes these levels across a light range: equal code steps can represent unequal light increments. It helps allocate precision for a task but does not create more stored codes.

The inverse function recovers a corresponding linear level only to the precision retained in the recording. It cannot recover information removed by quantisation, compression or clipping. Floating-point processing has different constraints from a final eight-bit file.

In y = x^γ, gamma is the exponent. A software menu labelled Gamma often describes a broader category: encoding. A power law, sRGB, camera log and HDR functions may appear in one list.

Always state the direction. For a simple display exponent of 2.4, decoding is L = V^2.4 and inverse encoding is V = L^(1/2.4). A label without direction and normalisation is not a complete operation.

OETF means opto-electronic transfer function: the relation from linear scene light to a nonlinear video signal. BT.709 specifies a piecewise function with a linear region near zero and a nonlinear branch above it. This describes signal formation, not the light-output response of a monitor.

Literal application of a reference OETF is not a full description of every camera pipeline. Recording mode, image preparation and creative choices can affect the signal. A standard name does not replace identifying the actual source.

EOTF means electro-optical transfer function: the relationship from an input video signal to linear display light. BT.1886 depends on the selected black luminance Lb and white luminance Lw. The 2.4 exponent is only part of the definition; normalisation and non-zero black also affect the result.

With ideal zero black the equation reduces to Lw · V^2.4. Non-zero black is not simply the same power curve with a constant added. The lab displays values calculated from the reference model, not measurements of your monitor.

OOTF links scene light to output image light. In a complete system it expresses overall tonal reproduction, including the intended rendering. A correct chain therefore need not be the identity Lout = Lin.

With consistent definitions, the OETF and EOTF can be composed into an overall relationship. An inverse OETF used for computation and the EOTF of a target display are not necessarily the same function. The invertible-pair lab on this page is deliberately separate from a complete television system.

The numbers 2.2 and 2.4 are not image-quality ratings. For the same normalised code, a simple 2.4 power yields less relative midtone light than a 2.2 power. This is a computational property, not a universal instruction to darken a master.

Choose a target that agrees with the delivery requirements, calibrated viewing path and viewing conditions. Surround affects perception; a random player or brightly lit room does not become the new reference merely because the result seems attractive.

Exact sRGB uses a piecewise function, not just a 2.2 power. Encoding linear C up to and including 0.0031308 uses 12.92 · C; above that it uses 1.055 · C^(1/2.4) − 0.055. The inverse branches at a code of 0.04045.

The difference from a power approximation near black matters in numerical operations. An exact transfer function still does not guarantee identical perception on arbitrary screens: colour management and viewing conditions matter too.

Log, PQ and HLG cannot be described by one exponent. Camera log requires its specific equation and normalisation. PQ defines a particular relationship to display luminance; HLG has its own OETF and system relationship to display. Copying one code value between them without conversion changes its meaning.

Nonlinear encoding does not automatically make data display-referred. ACEScct is a nonlinear working representation with defined primaries; encoding does not remove the need for output rendering.

In CST, treat Input Color Space and Input Gamma as two parts of the input description: the RGB colourimetric basis and its encoding. The output is described separately. Tone and gamut rendering are further operations, not evidence that changing Gamma alone is always sufficient.

In a managed project some transforms occur outside the visible node chain. Describe the source-to-output path before adding an explicit CST. The Gamma control in the primary wheels is an image adjustment; it does not redefine a file’s transfer function.

A transfer function uses normalised values, while a file may store them in a different integer range. Normalise the code range correctly before decoding the curve. A Video/Full black-level error cannot be fixed reliably by choosing another gamma.

The experiment’s 8 and 10 bits mean a teaching full-range quantiser of 0…255 or 0…1023. For the specified ten-bit BT.709 RGB and luma representation, reference black and white are 64 and 940. These are not universal boundaries for every channel and container.

A normalized code of 0.5 does not imply half the physical luminance without a decoding function. Under sRGB it represents approximately 0.214 linear light. Absolute luminance additionally requires a scale reference. Fifty percent on a code scale, Y = 0.5 and 50 cd/m² are not interchangeable.

The BT.1886 settings in the lab describe a calculated reference response; they do not change your browser screen’s emitted luminance. Connecting the calculation to a physical device requires a known, verified viewing path. Keep function, code range and units alongside any measurement.

With uniform rounding in full-range 8-bit encoding, neighbouring codes are separated by 1/255. After nonlinear decoding, their light increments differ across the scale. Bit depth increases sampling precision; it does not change primaries, white point or automatically increase the captured dynamic range.

When checking reversibility, distinguish numerical error from rounding loss. A matched continuous pair can recover its input almost exactly while the quantized version cannot. Re-encoding lost information does not create the missing intermediate values. The lab does not simulate chroma subsampling, codec compression or dithering.