Video encoding glossary — 194 terms with definitions

No terms match your filters. Try a different keyword or clear filters.

PSNR

Logarithmic measure of the per-pixel error vs the source. Easy to compute, weakly correlated with human perception above ~35 dB.

Psycho-visual redundancy

Information present in the signal but not perceived by the viewer — leveraged by perceptual quantization, chroma subsampling and psychovisual rate control.

QP (Quantization Parameter)

A per-block knob controlling quantization step size. Lower QP = finer quantization = higher quality and bitrate. Range typically 0–51 in H.264.

Quadtree partitioning

Recursive division of a coding unit into four equal squares. Used by HEVC; AV1 extends with rectangular splits.

Quantization

Dividing transformed coefficients by a step size and rounding — the single irreversible step in lossy video coding. All loss happens here.

Rate control

The encoder logic that allocates bits across frames and blocks to hit a chosen bitrate or quality target.

Rate-distortion

The fundamental tradeoff between bitrate and distortion. Plotting one against the other yields an RD curve — the language of codec comparisons.

rav1e

Rust-language AV1 encoder by Xiph/Mozilla. Strong determinism and safety; competitive quality on mid-speed presets.

RDO

The encoder process of choosing coding modes (block size, prediction, QP) by minimising a joint cost of bitrate and distortion.

Reference frame

A previously decoded frame held by the decoder, available as a source for inter-prediction. Codecs allow multiple references per frame.

Residual

The difference between the predicted block and the true block. This is what gets transformed, quantized and entropy-coded.

Resolution

The pixel dimensions of a video frame, e.g. 1920×1080 (FHD), 3840×2160 (4K UHD). Higher resolution means more detail and more bits to encode.

RGB

Additive color model storing red, green and blue components per pixel. Used for capture and display; video codecs internally work in YCbCr.

Run-length coding

Lossless coding that replaces a run of identical values with a (value, count) pair. Heavily used on the trailing zeros of quantized coefficients.

Sampling

Converting a continuous signal into a discrete sequence of values, in time (frame rate) and in space (resolution).

SAO

HEVC in-loop filter that adds small per-sample offsets to compensate for typical quantization errors near edges and band shifts.

Scalable coding

Coding a single bitstream so that multiple quality/resolution levels can be extracted from it — historically SVC for H.264, now LCEVC and AV1 SVC.

Scene-cut detection

Encoder logic that identifies hard cuts and places an I-frame there — cheaper than predicting across a discontinuity.

Shannon limit

The theoretical lower bound on the bitrate needed to represent a source losslessly — its entropy. Practical codecs only approach it.

Slice

A subset of macroblocks/CTUs that can be parsed independently of others in the same frame — for error resilience and parallel decoding.

Spatial redundancy

Correlation between neighbouring pixels inside a single frame. Exploited by intra-prediction and the DCT.

SSIM

A perceptual quality metric that compares structural patterns between source and encoded video. Better correlated with perception than PSNR.

Subjective testing

Lab-based protocols for measuring perceived video quality with human viewers. ITU-R BT.500 is the canonical methodology.

Subpixel

A fractional position between two pixel grid points, important for motion estimation (quarter-pel, eighth-pel) and font rendering.

Superblock (SB)

AV1's largest coding unit, 128×128. Subdivided recursively for prediction and transform.

SVT-AV1

Intel/Netflix-led production-grade AV1 encoder. The reason AV1 went mainstream in 2024–2026 — fast, scalable, good quality.

Telecine / pulldown

Technique to convert 24 fps film to 30 fps interlaced TV by repeating fields in a 3:2 pattern. Inverse telecine removes it.

Temporal redundancy

Correlation between successive frames. Exploited by motion estimation and inter-prediction — the single biggest source of video compression efficiency.

Tile

A rectangular region of a frame coded independently from other tiles. The HEVC/AV1 way to enable wavefront parallel encoding/decoding.

Tone mapping

Mapping an HDR signal of one peak brightness onto a display with a different peak — preserving highlight detail and overall look.

Transcoding

Decoding video from one codec and re-encoding it in another — or in the same codec at different bitrate/resolution.

Transfer function

A nonlinear mapping between scene light and stored code values. Examples — gamma 2.4 (SDR), PQ (HDR10/Dolby Vision), HLG (broadcast HDR).

Transform coding

Step of mapping pixel residuals into a frequency-domain representation before quantization.

Transmuxing

Repacking the same coded bitstream into a different container, without re-encoding. Fast, lossless and cheap.

Two-pass encoding

Encoding mode that runs an analysis pass first and an encode pass second — much better bitrate distribution than single-pass at the same average rate.

Variable frame rate (VFR)

Video where the interval between frames varies (e.g. screen recordings, some phone captures). Many encoders prefer CFR for stable rate control.

VBR

Rate-control mode that allows bitrate to vary by scene complexity, achieving better quality at the same average bitrate than CBR.

VideoToolbox

Apple's hardware video acceleration framework on macOS and iOS. Excellent power efficiency; H.264, HEVC, ProRes; AV1 decode from M3 Pro/Max onwards.

VMAF

Netflix's machine-learned perceptual quality metric on a 0–100 scale. The de facto standard for streaming-quality benchmarking.

VP8

Google's 2010 codec, open and royalty-free. Heavily used historically by WebRTC; now superseded by VP9 and AV1 for new deployments.

VP9

Google's 2013 codec, roughly on par with HEVC, royalty-free. The default delivery codec on YouTube for years; widespread browser support.

VPU

A dedicated chip designed solely for video transcoding (e.g. NETINT Quadra). Trades flexibility for stream density per watt.

vvenc / VVdeC

Fraunhofer's open-source VVC encoder (vvenc) and decoder (vvdec) — the practical entry point for H.266 experiments.

Wavefront parallel processing

HEVC's mechanism that lets multiple CTU rows be processed in parallel as long as each lags the previous one slightly. Standard parallelism trick.

WebM

Open container built on Matroska, restricted to royalty-free codecs (VP8/VP9/AV1 + Opus/Vorbis). Common in the Chromium/web ecosystem.

Widevine

Google's DRM, used on Android, Chrome, Cast and most non-Apple streaming devices. Three security levels (L1/L2/L3) gate access to HD/4K content.

x264

The dominant open-source H.264 encoder. Mature, tunable from preset=ultrafast to placebo, the reference for AVC quality benchmarks.

x265

The dominant open-source HEVC encoder, sibling to x264. The default starting point for any HEVC delivery pipeline.