Glossary of terms

The core vocabulary

Macroblock. A fixed square group of luminance samples, conventionally 16 by 16, treated as one unit for motion prediction and for the ordering of the coded data. Older standards build their whole picture from macroblocks; newer ones generalise the idea into a coding tree unit but keep the term for compatibility.

Group of pictures (GOP). A run of coded pictures that begins with an independently coded picture and continues through the predicted and bi-predicted pictures that depend on it. Its length fixes how often a stream offers a clean entry point.

I-frame, P-frame, B-frame. The three picture types. An I-frame is coded from its own data, a P-frame from an earlier reference, and a B-frame from a reference before and after it.

Discrete cosine transform (DCT). The reversible rewrite of a block of samples as a sum of cosine waves. It concentrates the smooth content of a block into few large values and the fine detail into many small ones, which is what makes later quantisation effective.

Quantisation. Division of a transformed value by a step, followed by rounding to an integer. The step controls how much detail is kept and how many bits are spent. Stated simply, it is where a lossy encoding stage actually chooses to lose something.

Luma and chroma. Luma is the brightness channel, written Y; chroma is one of the two colour-difference channels, written Cb and Cr. The separation lets the brightness be kept at full resolution while colour is reduced.

Chroma subsampling. The reduction of chrominance resolution relative to luminance, written as 4:4:4, 4:2:2 or 4:2:0. It saves raw data before any transform is applied.

Colour depth. The number of bits used for each sample, commonly eight, ten or twelve. Greater depth reduces banding in smooth gradients.

Bitrate. The number of bits carried each second, expressed in bits per second and its multiples. The average bitrate predicts the volume of a recording; the peak bitrate decides the buffer and the link.

Transport stream. A carriage that interleaves the packets of several elementary streams with timing marks, so that a receiver can recover the picture, the sound and the service information from one channel.

Further terms

Elementary stream. The raw coded output of one encoder, before it is sliced into packets.

Slice. A portion of a coded picture that can be decoded independently within the picture, used to bound the effect of a lost packet.

Motion vector. A pair of numbers describing where a matching patch was found in a reference picture.

Residual. The correction that remains after prediction, formed as the difference between the actual patch and the predicted one.

Reference picture. A previously coded picture whose samples may be used to predict a later one.

Frame rate. The number of pictures presented each second. It fixes the finest motion the signal can describe.

Nyquist rate. Twice the highest frequency present in a signal, the minimum sampling rate at which the signal can be reconstructed without error.

Aliasing. The folding of detail above the Nyquist rate onto lower frequencies, producing a pattern that was never present in the scene.

Colour gamut. The set of colours a given display or specification can reproduce, bounded by its three primaries.

Transfer function. The curve that relates stored code values to light, applied at capture and inverted at display so that dark tones are represented evenly.

Bits per pixel. The finished bitrate divided by picture size and frame rate. It normalises operating points of different sizes for comparison.

Profile. A defined subset of the features of a standard, so that a decoder need implement only what its role requires.

Buffer. A store that lets a variable-rate coded signal be carried over a fixed-rate channel by holding a short reserve of data.

Group length. The number of pictures between two independently coded pictures; used interchangeably with the length of a group of pictures.

Where a term has more than one accepted definition, this reference uses the one that serves the mathematics of the digital video signal, and notes the alternative in the body text where it matters. A reader who meets an unfamiliar abbreviation in a standard should first look for it here, and then follow the sheet that treats its subject in full.