Comparison of video codecs

The compression may employ lossy data compression, so quality measurement issues become important. Shortly after the compact disc became widely available as a digital-format replacement for analog audio, it became feasible to also store and use video in digital form. A variety of technologies soon emerged to do so. The primary goal for most methods of compressing video is to produce video that most closely approximates the fidelity of the original source and simultaneously deliver the smallest file size possible. However, there are also several other factors that can be used as a basis for comparison.

Introduction to comparison

The following characteristics are compared in video codecs comparisons:

• Video quality per bitrate (or range of bitrates). Commonly video quality is considered the main characteristic of codec comparisons. Video quality comparisons can be subjective or objective.
• Performance characteristics like compression/decompression speed, supported profiles/options, supported resolutions, supported rate control strategies etc.
• General software characteristics, for example:
  • Manufacturer
  • Supported OS (Linux, Mac OS, Windows)
  • Version number
  • Date of release
  • Type of license (commercial, free, open source)
  • Supported interfaces (VfW, DirectShow, etc.)
  • Price for codec (volume discounts, etc.)

Video quality

The quality the codec can achieve is heavily based on the compression format the codec uses. A codec is not a format, and there can be multiple codecs that implement the same compression specification – for example, MPEG-1 codecs typically do not achieve quality/size ratio comparable to codecs that implement the more modern H.264 specification. But quality/size ratio of output produced by different implementations of the same specification can vary, too.

Prior to comparing codec video quality, it is important to understand that every codec can give a varying degree of quality for a given set of frames within a video sequence. Numerous factors play a role in this variability. First, all codecs have a bitrate control mechanism which is responsible for determining the bitrate and quality on a per-frame basis. A difference between variable bit rate (VBR) and constant bit rate (CBR) creates a trade-off between a consistent quality over all frames, and a more constant bitrate, which is required for some applications. Second, some codecs differentiate between different types of frames such as key frames and non-key frames, differing in their importance to overall visual quality and the extent to which they can be compressed. Third, quality depends on prefiltrations, that is included on all present-day codecs. Other factors can also come into play.

For a sufficiently long clip, it is possible to select sequences which have suffered little from the compression and sequences which have suffered heavily, especially if CBR was used, in which the quality between frames can vary highly due to different amounts of compression needed to achieve a constant bitrate. So, in any one long clip such as a full length movie, any two codecs may perform quite differently on a particular sequence from the clip, while the codecs may be approximately equal (or the situation reversed) in quality over a wider sequence of frames. Press-releases and amateur forums sometimes select sequences known to favor a particular codec or style of rate control in reviews.

Objective video quality

Objective video evaluation techniques are mathematical models that approximate results of subjective quality assessment, but are based on criteria and metrics that can be measured objectively and automatically evaluated by a computer program. Objective methods are classified based on the availability of the original video signal, which is considered to be of high quality (generally not compressed). Therefore, they can be classified as:

• Full reference methods (FR), where the whole original video signal is available
• Reduced reference methods (RR), where only partial information of the original video is available, and
• No-reference methods (NR), where the original video is not available at all.

The main FR metrics are:

Peak signal-to-noise ratio (PSNR)

The most widely used video quality metric during the last 20 years (used approximately in 99% of scientific papers and in 20% of marketing materials). However, the validity of this metric is limited. It is only conclusive when the same codec (or codec type) and content is being compared.^[1][2]

Structural similarity (SSim.)

A new metric (suggested in 2004) which shows better results than PSNR at the cost of a reasonable increase in computational complexity.

VQuad-HD an ITU-T J.341 standard

The new standard was recently (Jan 2011) approved by ITU-T as J.341. The new VQuad-HD™ algorithm was developed by Swissqual in 2008-2010. It was the best performing model in the HDTV competition to find the new standard that was organized by the independent and non-commercial Video Quality Expert Group (VQEG).More information on VQuad-HD can be found in the technical white paper "Video Quality Measurement for High Definition Video Signals" available for download from: white paper

Some other metrics have been suggested by Video Quality Experts Group (VQEG), private companies, and universities, but are not widespread.

The main comparison method is the so-called RD-curve (rate/distortion chart), where a metric value is plotted against the Y-axis and the bitrate against the X-axis.

Some example NR metrics are:

• Blocking measure — measurement power of so called blocking artefacts (extremely noticeable without deblocking filter usage on low bitrates)
• Blurring measure — measurement of common video blurring (washout)

Subjective video quality

This is concerned with how video is perceived by a viewer and designates his or her opinion on a particular video sequence. Subjective video quality tests are quite expensive in terms of time (preparation and running) and human resources.

There is an enormous number of ways of showing video sequences to experts and of recording their opinions. A few of them have been standardized. They are thoroughly described in ITU-R recommendation BT.500.

Following subjective video quality comparison methods are used:

• Double Stimulus Impairment Scale (DSIS) — suggested in ITU-R BT.500-11.
• Double Stimulus Continuous Quality Scale (DSCQS) type I and type II — suggested in ITU-R BT.500-11
• Stimulus Comparison Adjectival Categorical Judgement (SCACJ) — suggested in ITU-R BT.500-11
• Subjective Assessment Method for Video Quality evaluation (SAMVIQ)
• MSU Continuous Quality Evaluation (MSUCQE)

The reason for measuring subjective video quality is the same as for measuring the Mean Opinion Score for audio. Opinions of experts can be averaged; average mark is usually given with confidence interval. Additional procedures can be used for averaging, for example experts who give unstable results can be rejected (for instance, if their correlation with average opinion is small).

In case of video codecs, this is a very common situation. When codecs with similar objective results show results with different subjective results, the main reasons can be:

• Pre- and postfilters are widely used in codecs. Commonly codecs use prefilters like video denoising, deflicking, deshacking, etc. Denoising and deflicking commonly maintain PSNR value, but increase visual quality (the best slow denoising filters also increase PSNR on middle and high bitrates). Deshacking seriously decreases PSNR, but increases visual quality. The same situation with postfilters — deblocking and deringing maintain PSNR, but increase quality. Graining (suggested in H.264) essentially increases video quality especially on big plasma screens, but decrease PSNR.

Note: All filters worsen compression/decompression time, so they increase visual quality, but decrease speed.

• Motion estimation (ME) search strategy can also cause different visual quality for the same PSNR. So called true motion search commonly will not reach minimum sum of absolute differences (SAD) values in codec ME, but may result in better visual quality. Also such methods require more compression time.
• Rate control strategy. VBR commonly cause better visual quality marks than CBR for the same average PSNR values for sequences.

It is difficult to use long sequences for subjective testing. Commonly, three or four ten-second sequences are used, compared with full movies used for objective metrics. Sequence selection is important — those sequences that are similar to the ones used by developers to tune their codecs are more competitive.

Performance comparison

Speed comparison

Number of frames per second (FPS) commonly used for compression/decompression speed measurement.

The following issues should be considered when estimating probable codec performance differences:

• Decompression (sometimes compression) frame time uniformity. Big differences in this value can cause annoyingly jerky playback.
• SIMD support by processor and codec — e.g., MMX, SSE, SSE2, each of which change CPU performance on some kinds of tasks (often including those with which codecs are concerned).
• Multi-threading support by processor and codec (sometimes turning on Hyper-threading support (if available on a particular CPU) causes codec speed to decrease)
• RAM speed (generally important for most codec implementations)
• Processor cache size (low values sometimes cause serious speed degradation, e.g. for CPUs with low cache such as several of the Intel Celeron series.)
• GPU usage by codec — some codecs can drastically increase their performance by taking advantage of GPU resources.

So, for example, codec A (being optimized for memory usage, i.e. uses less memory) may give slower performance on modern computers (which are typically not memory limited) than codec B. The same pair of codecs may give opposite results if running on an older computer with reduced memory (or cache) resources.

Profiles support

Modern standards define a wide range of features and require very substantial software or hardware efforts and resources for their implementation. Only selected profiles of a standard are typically supported in any particular product. (This very common situation for H.264 implementations for example.)

The H.264 standard includes the following seven sets of capabilities, which are referred to as profiles, targeting specific classes of applications:

• Baseline Profile (BP): Primarily for lower-cost applications with limited computing resources, this profile is used widely in videoconferencing and mobile applications.
• Main Profile (MP): Originally intended as the mainstream consumer profile for broadcast and storage applications, the importance of this profile faded when the High profile was developed for those applications.
• Extended Profile (XP): Intended as the streaming video profile, this profile has relatively high compression capability and some extra tricks for robustness to data losses and server stream switching.
• High Profile (HiP): The primary profile for broadcast and disc storage applications, particularly for high-definition television applications (this is the profile adopted into HD DVD and Blu-ray Disc, for example).
• High 10 Profile (Hi10P): Going beyond today's mainstream consumer product capabilities, this profile builds on top of the High Profile — adding support for up to 10 bits per sample of decoded picture precision.
• High 4:2:2 Profile (Hi422P): Primarily targeting professional applications that use interlaced video, this profile builds on top of the High 10 Profile — adding support for the 4:2:2 chroma sampling format while using up to 10 bits per sample of decoded picture precision.
• High 4:4:4 Predictive Profile (Hi444PP): This profile builds on top of the High 4:2:2 Profile — supporting up to 4:4:4 chroma sampling, up to 14 bits per sample, and additionally supporting efficient lossless region coding and the coding of each picture as three separate color planes.
• Multiview High Profile: This profile supports two or more views using both inter-picture (temporal) and MVC inter-view prediction, but does not support field pictures and macroblock-adaptive frame-field coding.

The standard also contains four additional all-Intra profiles, which are defined as simple subsets of other corresponding profiles. These are mostly for professional (e.g., camera and editing system) applications:

• High 10 Intra Profile: The High 10 Profile constrained to all-Intra use.
• High 4:2:2 Intra Profile: The High 4:2:2 Profile constrained to all-Intra use.
• High 4:4:4 Intra Profile: The High 4:4:4 Profile constrained to all-Intra use.
• CAVLC 4:4:4 Intra Profile: The High 4:4:4 Profile constrained to all-Intra use and to CAVLC entropy coding (i.e., not supporting CABAC).

Moreover, the standard now also contains three Scalable Video Coding profiles.

• Scalable Baseline Profile: A scalable extension of the Baseline profile.
• Scalable High Profile: A scalable extension of the High profile.
• Scalable High Intra Profile: The Scalable High Profile constrained to all-Intra use.

An accurate comparison of codecs must take the profile variations within each codec into account.

See also MPEG-2 Profiles and Levels.

Supported rate control strategies

Videocodecs rate control strategies can be classified as:

• Variable bit rate (VBR) and
• Constant bit rate (CBR).

Variable bit rate (VBR) is a strategy to maximize the visual video quality and minimize the bit rate. On fast motion scenes, a variable bit rate uses more bits than it does on slow motion scenes of similar duration yet achieves a consistent visual quality. For real-time and non-buffered video streaming when the available bandwidth is fixed, e.g. in videoconferencing delivered on channels of fixed bandwidth, a constant bit rate (CBR) must be used.

CBR is commonly used for videoconferences, satellite and cable broadcasting. VBR is commonly used for video CD/DVD creation and video in programs.