Fernando Voltolini de Azambuja

Imaging and color measurement

CAM16 and Hellwig–Fairchild calculator

Compare what standard CAM16 and the 2022 Hellwig–Fairchild proposal predict for the same XYZ stimulus and viewing conditions. The calculation runs in your browser.

Inputs

Values are never guessed. Use one scale for the stimulus, adopted white, and background; adapting luminance remains absolute.

Stimulus XYZ
Adopted white XYZ

Normalization never changes L_A; it is an absolute luminance in cd/m².

Results

Appearance correlates

Reference result for the example inputs.

CorrelateCAM162022 proposal
Lightness J59.130559.1305
Brightness Q232.56179.1392
Chroma C49.866434.4892
Colorfulness M51.835345.5003
Saturation s47.211157.494
Hue angle h45.6445.64

Shaded rows are redefined by the proposal. The columns use different scales, so their raw magnitudes are not a color difference and are not directly rankable.

Hue is shown only when the opponent response is large enough to distinguish a direction from floating-point cancellation.

Model implementation 1.2.1

What the models report

Both models describe appearance with lightness J, brightness Q, chroma C, colorfulness M, saturation s, and hue angle h. The proposal preserves J and h while changing the definitions and scales of the other four correlates.

These outputs are not display RGB, a color-difference score, a measurement, or a verdict about which model predicts observers better.

Why viewing conditions matter

The same XYZ can appear different as the adopted white, background, surround, or adapting luminance changes. A plausible-looking result under the wrong condition is still the wrong calculation, so every condition stays visible and editable.

Why compare the formulations

The 2022 proposal revisits linked brightness, chroma, colorfulness, and saturation relations in CAM16. Its published fits improve for brightness and chroma but decline for colorfulness—a mixed result, not a universal replacement. See the equation study or use the Python and JavaScript source.

The public suite runs on Windows, macOS, and Linux and compares both models against an independent implementation over 1,512 declared conditions.