High Efficiency Video Coding

H.265, also known as HEVC (High Efficiency Video Coding, 高效率视频编码), is a video compression standard jointly developed by ITU-T and ISO/IEC. Experts from both organizations formed the Joint Collaborative Team on Video Coding (JCT-VC) in 2010, and the first version was approved in 2013, designated as ITU-T H.265 and ISO/IEC 23008-2, respectively; Within the MPEG standards framework, it is also known as MPEG-H Part 2.

H.265 and HEVC are two names for the same standard.

The former derives from the ITU-T Recommendation number, while the latter is the standard’s abbreviation.

HEVC retains the hybrid coding framework consisting of intra-frame prediction, motion compensation, transformation, quantization, and entropy coding, but has redesigned the image partitioning, prediction, and parallel processing mechanisms to accommodate various applications ranging from low-resolution video to ultra-high-definition programming. HEVC no longer uses fixed 16×16 macroblocks as the top-level image unit. The frame is first divided into encoding tree units, whose luminance regions in the first version often reached 64×64 sample points, and then recursively subdivided into encoding units according to a quadtree structure. A large, flat area can be represented by a small number of large blocks, while object edges, text, or complex motion can be handled using smaller partitions. Below the coding units, there are also prediction units used to describe intra- or inter-frame prediction regions, as well as transform units that carry the residual transform structure; Consequently, prediction and transform boundaries do not need to coincide exactly. Intra-frame prediction offers planar, DC, and 33 angular modes to infer the current block from already reconstructed neighboring samples. The increased number of directions helps describe diagonal lines, contours, and regular textures, but also presents the encoder with more candidate solutions. Interframe prediction can utilize multiple reference frames, motion vectors with one-quarter luminance sampling precision, merge modes, and advanced motion vector prediction; the standard defines how to represent and decode this information but does not specify which search algorithm the encoder must use.

The size of the residual transform can be expanded from 4×4 to 32×32. Larger transforms are suitable for areas with gradual changes, while smaller transforms are used for details and edges; quantization remains the primary step in lossy compression that results in information loss. HEVC uniformly uses CABAC for entropy coding and does not retain the option to use CAVLC alongside CABAC, as in H.264. Loop processing includes deblocking filtering and adaptive sample-point offset: the former reduces discontinuities at block boundaries, while the latter applies an offset based on the category of the reconstruction sample to correct certain quantization errors. The processed frame is then fed into the decoding frame buffer and used for subsequent prediction.

To support multi-core processing and hardware pipelining, HEVC specifies parallel processing of slices, blocks, and wavefronts. Blocks divide the frame into approximately independent rectangular regions, suitable for parallel decoding or local access; wavefront parallel processing allows adjacent encoding tree units to start interleaved after satisfying minimal dependencies. Parallel boundaries may reduce some cross-block prediction or context inheritance; excessive use may lower compression efficiency, so the encoder must balance speed, latency, error isolation, and bitrate.

The bitstream is organized into network abstraction layer units, and decoding configurations are passed through video parameter sets, sequence parameter sets, and picture parameter sets. HEVC also distinguishes between random-access pictures such as IDR, BLA, and CRA: these have different semantics regarding where decoding begins, whether previous pictures can continue to serve as references, and how open picture groups are spliced together; they cannot be uniformly replaced by the term “key frame.” The display order may also differ from the decoding order.

The first version defined three main profiles: Main, Main 10, and Main Still Picture. Main is intended for 8-bit 4:2:0 video; Main 10 extends the allowed bit depth to 10 bits; and Main Still Picture is used for single images. Subsequent scope extensions added support for higher bit depths, 4:2:2, 4:4:4, RGB, intra-frame-only, and professional production capabilities, along with extensions for scalable, multi-view, 3D video, and screen content encoding. Levels restrict frame size, sampling rates, and buffer resources; Main Tier and High Tier specify different maximum bitrates for certain levels. Detecting HEVC alone does not necessarily mean the video is 10-bit, 4K, or HDR.

HEVC is a video compression standard, not a container format.

It can be encapsulated in MP4, QuickTime, Matroska, MPEG-2 Transport Stream, and M2TS, or saved as a bare bitstream. Containers manage sample time, tracks, audio, subtitles, and file-level metadata; video color profiles and HDR metadata also involve corresponding stream and application specifications. Changing the file extension from MKV to MP4 does not alter the internal HEVC encoding content. Ultra HD Blu-ray uses HEVC Main 10 to carry a main video stream with a resolution of up to 3840×2160 and 10-bit color depth; broadcasting and streaming services also commonly use HEVC to deliver ultra-high-definition or high-dynamic-range programming. However, HEVC itself can also encode standard-definition, 1080p, 8-bit, or standard dynamic range video. The BDMV specification for standard Blu-ray differs from the ROM4 specification for Ultra HD Blu-ray; the former supports AVC, VC-1, and MPEG-2 Video, and a player’s ability to play certain HEVC files does not automatically mean it is compatible with UHD discs.

“Approximately half the bitrate for the same picture quality” is a generalization describing HEVC’s design goals and specific test results; it is not a fixed ratio that applies to all source material and settings. Resolution, noise, motion complexity, encoder implementation, search depth, and speed presets can all alter the comparison results. x265 is a common HEVC software encoder; hardware encoders and other software can also generate compliant bitstreams. Implementation names should not be used in place of H.265 / HEVC as the official name of the coding standard.