Reference
194 terms with definitions, aliases, and cross-references to deep-dive articles. Search by name, filter by chapter, or jump straight to a letter.
No terms match your filters. Try a different keyword or clear filters.
Predicting a block from blocks in previously decoded reference frames, exploiting temporal redundancy via motion vectors.
I
Coding where every frame is intra — like a sequence of still images. High bitrate, but each frame is independently editable.
Predicting a block from neighbouring already-decoded blocks in the same frame, exploiting spatial redundancy.
Visible motion jerkiness caused by mismatched frame rates between source and display (e.g. 24 fps cinema on a 60 Hz TV without pulldown).
J
Units of bitrate — thousands and millions of bits per second. 1080p H.264 streaming sits around 4–8 Mbps; 4K HDR around 15–25 Mbps.
K
Informal term for an I-frame or IDR — a frame at which playback can start.
J = D + λ·R — combined cost of distortion D and rate R, weighted by λ. The standard objective minimised by RDO.
L
Enhancement layer that sits on top of any base codec (H.264/HEVC/AV1), adding residual detail at higher resolution. Improves quality without changing the base bitstream.
A named upper bound on resolution/bitrate/buffer that a decoder must support — e.g. H.264 Level 4.1 = 1080p30 @ ~10 Mbps.
The AOMedia reference AV1 encoder. Highest quality on slow presets, slowest of the AV1 encoders; benchmark baseline.
Google's reference VP8/VP9 encoder. Was the YouTube workhorse for years; still common in WebRTC pipelines.
Reconstructs the input bit-for-bit. Used in mastering and archive (FFV1, AVC-Intra). Compression ratios typically 2–4:1 — too low for streaming.
Discards information the eye is unlikely to notice. All practical video delivery codecs (H.264, HEVC, AV1) are lossy.
The brightness component of a YCbCr pixel. Gamma-corrected (hence the prime). Carries most of the perceptual information.
The 16×16 coding unit of H.264 and earlier codecs. HEVC replaced it with the variable-sized CTU, AV1 with the superblock.
M
HEVC profile supporting 4:2:0 video up to 10-bit. Mandatory for HDR HEVC delivery and the default modern profile for streaming.
The visual system's reduced sensitivity to errors in busy or moving regions. Used by encoders to allocate fewer bits there.
Megabits per second. Standard unit for video bitrate at HD/UHD scale.
Open and flexible container, popular for high-quality private libraries. Supports virtually any codec and many subtitle/audio tracks.
Selecting which prediction mode and block partition to use for each region of the frame — driven by RDO.
An interference pattern produced when a fine repetitive pattern (fabric, fence) clashes with the pixel grid — a classic aliasing artifact.
Average subjective quality score (typically 1–5) given by a panel of viewers. The ground truth that every objective metric tries to approximate.
Smear of moving objects within one frame, caused by the camera shutter being open while the object moves. Critical to cinematic look.
The encoder's search for the best-matching block in a reference frame. By far the most expensive step in encoding.
A 2D offset (often at sub-pixel precision) saying "this block looks like that block in the reference frame, shifted by (Δx, Δy)".
Apple's QuickTime container, the historical ancestor of MP4. Common in production pipelines (ProRes, DNxHR).
The dominant delivery container, built on ISO Base Media File Format. Used everywhere on the web; carries H.264/HEVC/AV1.
1995 codec, the foundation of digital TV, DVB, ATSC and DVD. Replaced for delivery by H.264 but still alive in broadcast and DVB legacy.
Transport Stream container, originally for broadcast — small, packet-based, recoverable from packet loss. The legacy segment format for HLS.
Mean of squared per-pixel differences between source and encoded video. PSNR is just a logarithmic transform of MSE.
Multi-scale extension of SSIM, evaluating quality at several spatial resolutions. Standard in modern codec benchmarks.
A codec whose entire compression pipeline is a neural network, trained end-to-end. Promising on perceptual metrics; not yet practical for general delivery.
N
NVIDIA's hardware video encoder, on every modern GeForce/Quadro/Tesla. Encodes H.264, HEVC and AV1 at multi-stream realtime speeds.
To represent a signal of bandwidth B without loss, sample it at ≥ 2B. Violating it causes aliasing — the root cause of moiré and stairstep artifacts.
GOP whose final B-frames reference the next GOP's I-frame. Slightly smaller; awkward for ABR segment boundaries.
O
Splitting encoding work across CPU cores, GPUs or machines — via tiles, slices, frame-parallel or chunked encoding.
P
Adjusting quantization per region/frequency based on visual perception — quantize harder where the viewer won't notice (textures, motion).
A more granular extension of per-title — different scenes inside one title get different bitrate/QP. Even bigger savings.
Custom bitrate ladder per video, based on its complexity. Pioneered by Netflix; saves 20–35 % bitrate at the same quality.
A frame predicted from one or more previous frames. Much smaller than an I-frame, the workhorse of compression.
The smallest addressable element of a digital image. Each pixel stores intensity and color, usually as three or four numeric components.
The shape of an individual pixel — square (1:1, default for most modern video) or non-square (legacy NTSC/PAL, anamorphic capture).
Microsoft's DRM, used on Edge, Xbox, smart TVs and many set-top boxes.
Absolute-luminance HDR transfer function up to 10 000 cd/m². Used by HDR10, HDR10+ and Dolby Vision. Display-referred, requires metadata.
General term for "encoder's best guess for a block before coding the residual" — comes in intra and inter flavours.
A named subset of codec features required by a decoder, e.g. H.264 Baseline / Main / High, HEVC Main / Main10.
Modern scan mode where each frame is transmitted as one complete image. Default for streaming, web, gaming and most production.
Apple's intra-only production codec. High bitrate, fast decode, great for editing — not for delivery.
Specialist software house for video, real-time and AI products. Founded 2005. 50 in-house engineers.
By submitting data in this form, you agree with the Personal Data Processing Policy.