Scene-Referred vs. Display-Referred (Output-Referred) Imagery
Most production images are either scene-referred or display-referred.
Quote
"… categorize color spaces by the 'direction' of this relationship to real world quantities, which we refer to as image state. Color spaces which are defined in relation to display characteristic are called display-referred, while color spaces which are defined in relation to input devices (scenes) are scene-referred. While there are other flavors of images states (intermediate-referred, focal-plane referred) display-referred and scene-referred colorimetry are most commonly used in motion-picture color management …"
Jeremy Selan, Sony Pictures Imageworks, Cinematic Color, VES, 2012
Display-Referred Imagery
Display-referred values describe output on a target display, not real-world exposure. Web, phone, television, and cinema images are display-referred and require no additional viewing transform on their intended display.
Almost all reference displays have a built-in gamma encoding. These encodings are different depending on the display, its usage and intention, and the assumed viewing environment. For example, an sRGB image on a computer monitor or phone display is viewed at roughly gamma 2.2, home video Rec.709 material through a display EOTF of about gamma 2.4 (BT.1886), and standard dynamic range digital cinema through gamma 2.6.
In analog terms, a display-referred image is our film print.
An image intended for exhibition will ultimately be display-referred.
Scene-Referred Imagery
Scene-referred values preserve relative exposure relationships in a photographed or CG scene. They do not describe final display output or absolute light.
Scene-referred images in the context of digital cinema workflow are any camera original images (or visual effects shots encoded in camera original encoding) that retain the camera's native dynamic range and tonal response.
Camera RAW images are inherently scene-referred, while traditional video images are display-referred.
Scene-referred images can come in two major flavors: scene-linear (linear light space) or any number of camera log encoding schemes (such as ARRI LogC3 / LogC4, RED Log3G10, Sony S-Log3, Canon Log 2 / Log 3, Blackmagic Film Gen 5, or Apple Log). Always record the version number. "LogC" and "Canon Log" each name more than one curve, and they are not interchangeable.
Scene-Linear
Scene-linear encoding is a purely linear (no gamma) encoding of relative light values, expressed in floating point values. Middle gray is mapped to 0.18, each stop doubles the value, and highlights may exceed 1.0. Scene-linear images are commonly written to half-float OpenEXR files, 16-bit floating point, IEEE 754 binary16, rather than historically integer-based formats like DPX or QuickTime. The distinction matters: "16-bit" on its own is ambiguous in post, where it can mean either half-float or 16-bit integer, and only the float type carries values above 1.0.
Scene-linear is not to be confused with normalized linear, or camera linear, which are float or integer encoded values ranging from 0.0 to 1.0 and correspond directly to a camera's sensor raw analog-to-digital output values prior to encoding.
Camera Log
Log encoding expresses scene exposure in stops, preserving wide dynamic range in 10- or 12-bit integer files. Scene-linear work generally requires 16-bit floating point.
As the dynamic range of new cameras increases, the need for 16-bit image encoding increases. 10-bit log is insufficient for incredibly high dynamic range encodings, as it can produce quantization artifacts (or banding) if too much dynamic range is compressed into a small range of code values. This is particularly problematic when mastering for HDR, and for theatrical delivery, where the projected image carries more range than a Rec.709 video master. In those situations 12-bit log should really be the minimum bit depth of camera original images, with 16-bit log or 16-bit scene-linear being ideal.
The major reason we still use logarithmically encoded images is that it allows us to capture scene-referred imagery to a variety of convenient formats without the overhead of Camera RAW. The ARRI ALEXA 35 and RED V-RAPTOR allow for native ProRes capture, while Sony's VENICE 2 and BURANO allow native capture to other raster image formats like XAVC. Not every camera offers this. Blackmagic's URSA Cine line, for instance, records only Blackmagic RAW internally (alongside H.264 proxies), so it remains a RAW workflow whether you want one or not. External recorders provide further options for capturing scene-referred imagery without Camera RAW.
As Camera RAW typically exhibits higher data rates than ProRes and necessitates more intensive dailies processing, Camera RAW may be an expensive luxury for some productions. In particular, broadcast television productions rarely utilize the full benefits of RAW and instead record in a camera log based format like ProRes at a convenient resolution and format.
Camera log images are analogous to log film scans and fit in with established workflows very easily.
Video ≠ Linear
Do not use linear as shorthand for video. Rec.709 video is display-referred and gamma-encoded. It is not linear.
The distinction matters in VFX specifications. “Linear plates” normally means scene-linear OpenEXR, not display-referred video.
Example
When a client tells you they are providing "linear plates" you would naturally expect a scene-linear, scene-referred workflow centered around OpenEXRs. But if they really mean video, not linear, there are some serious ramifications for your visual effects and color grading workflow.