SHEET 03 · Coding standards
Coding standard comparison
A side-by-side account of four widely documented coding standards — ISO/IEC MPEG-2, ITU-T H.264 AVC, ITU-T H.265 HEVC and AOMedia AV1 — across block structure, colour precision and efficiency.
How to compare coding standards honestly
Two coding standards can only be compared on a common footing. The honest approach is to fix the picture, the frame rate and a quality measure, then record the number of bits each scheme needs to reach that quality. In practice the published literature reports a range rather than a single figure, because the result depends on the content, the encoder and the configuration of parameters. The table below therefore gives representative ranges rather than promises, and each row should be read as a family of behaviour rather than a fixed value.
| Standard | Issuing body | Block structure | Colour depth (bit) | Typical throughput | Efficiency versus MPEG-2 |
|---|---|---|---|---|---|
| ISO/IEC MPEG-2 (H.262) | ISO/IEC MPEG + ITU-T | 16×16 macroblock, 8×8 transform | 8 | up to 80 Mbit/s | 1.0× (baseline) |
| ITU-T H.264 AVC | ITU-T + ISO/IEC MPEG | 16×16 macroblock, sub-partitions to 4×4 | 8–10 | up to 240 Mbit/s | about 2.0× |
| ITU-T H.265 HEVC | ITU-T + ISO/IEC MPEG | coding tree unit 64×64 down to 8×8 | 8–12 | up to 960 Mbit/s | about 3.0–4.0× |
| AOMedia AV1 | Alliance for Open Media | superblock up to 128×128, recursive split | 8–12 | up to 1 200 Mbit/s | about 3.0–4.5× |
The macroblock and its successors
MPEG-2 organises each picture into 16 by 16 macroblocks and transforms each block of 8 by 8 values. The arrangement is simple, regular and cheap to decode, which is why it served terrestrial broadcast for so long and still appears in legacy streams. Its limitation is that a fixed block size cannot adapt to the content: a smooth sky and a detailed crowd are treated with the same grid.
H.264 AVC keeps the macroblock frame but allows motion prediction and residual coding to be carried out on smaller partitions, down to 4 by 4 values. It also introduces an in-loop deblocking filter that smooths the boundaries between blocks, a refinement that removes much of the visible blocking at low bitrates. The combination of flexible partitioning and filtering accounts for the roughly twofold gain that published studies report over MPEG-2.
H.265 HEVC abandons the fixed macroblock in favour of a coding tree unit that may be as large as 64 by 64 values and may be split recursively into smaller units. A picture with large flat regions therefore spends very few units, while a detailed region subdivides until the prediction is good. This adaptivity, together with richer prediction modes and a sample-adaptive filter, produces the further gain of roughly 40 to 50 per cent over H.264 AVC.
AOMedia AV1 follows a similar philosophy with a superblock of up to 128 by 128 values and a wide set of techniques for prediction and filtering. It adds a loop restoration filter and a film-grain synthesis stage that models the noise of the original so that the encoder need not spend bits preserving it literally. Its efficiency sits in the same broad band as HEVC, but the standard was designed with delivery over packet networks in mind, which shows in how its parameters are arranged.
Colour precision
The colour depth column deserves a note. MPEG-2 is confined to eight bits per sample. H.264 AVC, H.265 HEVC and AV1 provide profiles that carry ten, twelve or even more bits per sample. Extra precision reduces the banding that appears in smooth gradients, and it gives the encoder more headroom before quantisation error becomes visible. The cost is greater raw data, but only for the profile in use; a ten-bit stream is not twice as large as an eight-bit one, because the compression stage reshapes the difference.
Plate 6 — representative relative efficiency. Values are ranges reported in the public literature and depend on content and configuration.
Practical consequences
The choice of standard rarely comes without penalty. A newer scheme demands more arithmetic in both encoder and decoder, and although hardware is built to match it, older receiving equipment may not support it. A broadcaster with a large installed base therefore often carries a dual stream, and the standards themselves define profiles that allow a single decoder to handle a subset of the features. For scientific study, the useful lesson is that each generation buys its gain by describing the picture more precisely — smaller effective units, richer prediction and a better model of what the eye will tolerate — and never by promising something for nothing.