Glossary of video compression terms

Reference · 00:00:13 · reviewed 2026-10-02

Short, precise definitions for the vocabulary used across this portal. Terms are grouped by theme so that related concepts appear together.

Structure and signalling

Codec — the algorithm that compresses and reconstructs video, audio or both. The codec determines how few bits are needed for a given quality.
Container — the envelope that indexes, times and describes the streams inside a media file. It does not define compression.
Bitstream — the compressed sequence of bits produced by an encoder and consumed by a decoder.
Elementary stream — a single compressed track, such as one video track or one audio track, before it is placed in a container.
Muxing — combining elementary streams into a container. Its inverse, demuxing, separates them again.
Profile — a named set of coding tools that a decoder must support to decode a stream.
Level — a named resource envelope bounding resolution, frame rate, bitrate and buffer size for a decoder.

GOP — group of pictures, the span from one keyframe up to the frame before the next keyframe.
I-frame — a self-contained frame coded without reference to any other, usable as a seek point.
P-frame — a frame predicted from earlier frames in display order.
B-frame — a frame predicted from frames on both sides of it, which requires reordering between decode and display.
Keyframe interval — the distance between successive seek points, usually expressed in frames or seconds.
Decoder reordering — the process that restores presentation order from the decode order implied by B-frames.

CRF — constant rate factor, a perceptual quality target on a per-codec scale where lower values mean higher fidelity.
RF — rate factor, the analogous quality control in some AV1 encoders, on a comparable but distinct scale.
Bitrate — the number of bits consumed per unit of time, commonly averaged over the whole file or measured in short windows.
Capped quality — constant-quality encoding combined with a maximum bitrate ceiling to bound transport peaks.
Quantisation — the division of transformed coefficients by a step size, the dominant source of perceptual quality loss.
Psychovisual tuning — rate-control adjustments that spend bits where the eye is most sensitive.
Per-title encoding — analysing each title to choose rungs on its own efficient rate-quality frontier.

PSNR — peak signal-to-noise ratio, a simple fidelity metric that weights all errors equally.
SSIM — structural similarity, a metric weighting structural degradation more than uniform error.
VMAF — a fusion metric combining several signals into one perceptual score, requiring a reference pass.
Reference — the undistorted source against which a distorted encode is compared during measurement.

Chroma subsampling — storing colour difference channels at lower resolution than luma, commonly 4:2:0 for delivery.
Bit depth — the number of bits per sample; ten-bit reduces banding in smooth gradients compared with eight-bit.
Transfer function — the mapping between stored code values and light output, central to high dynamic range delivery.
HDR — high dynamic range, a wider tonal range than standard dynamic range, requiring metadata to reproduce faithfully.
Banding — visible steps in a smooth gradient, usually caused by insufficient bit depth or coarse quantisation.

Adaptive bitrate — delivery in which the client selects among several representations according to measured throughput and buffer.
Bitrate ladder — the set of representations offered to an adaptive client, spaced so that switching is gradual.
Segment — a short self-contained interval of media used by streaming delivery, whose length trades efficiency against latency.
Fast start — placing index information early in a file so playback begins before the whole file arrives.
Remux — rewriting the container around an unchanged payload, a lossless operation when done correctly.

If a term you need is missing, suggest it through the editorial desk and it will be considered for the next revision of this page.