H.264 vs H.

265
Malcolm Weir
George Nelson
Gary Thom
Delta Information Systems, Horsham, PA, USA
Ampex Data Systems Corporation, Hayward, CA, USA

Abstract: This paper discusses the application of two of times the number of pixels per second required by
the latest video encoding approaches for test platforms, 1080p30, then that same mechanism could be applied to
particularly in bandwidth-constrained and over imperfect less demanding formats (e.g. 1080p30) and achieve some
(lossy) transmission links. High Efficiency Video Coding level of bandwidth savings.
(HEVC or H.265) is compared with MPEG-4 Advanced It is this second rationale that has led to a category of
Video Coding (AVC or H.264), and the behavior of both in applications for H.265/HEVC that is totally distinct from
low bandwidth and lossy transmission channels is mass content distribution (as implemented by the likes of
explored, concentrating on the ability to transmit usable Netflix for UHD-TV): video transmission over constrained
information over low bandwidth links using each approach. network links, as with Apple’s FaceTime application
Comparison of the same video simultaneously compressed connecting iPhone 6 (and later) devices.
heavily for constrained links and lightly for on-board Therefore, with H.265/HEVC in use within a ubiquitous
storage is used to illustrate the effects of heavy mass-market smartphone, it is clear that here is a
compression on video usability, as well as side-by-side technology that may be ready for use in test and evaluation
comparison of the output of both video encoding applications.
algorithms as implemented in a rugged airborne package.
In the related market of airborne ISR technology, the US
Keywords: Video, Encoding, CODEC Department of Defense’s Motion Imagery Standards Board
(MISB), and the related STANAG (STANAG 4609) has
provisionally endorsed the use of H.265/HEVC. As the
1. Introduction entity responsible for ensuring interoperability of video
systems, this provides a strong indication that the
The latest video encoding standard, High Efficiency Video technology is stable and mature.
Coding (HEVC), from the body1 that developed the
H.264/MPEG-4 AVC standard was published in June
2. Commercial Aspects of H.265
2013. As with its predecessor, the standard has two names
reflecting the two standards bodies: ITU-T’s H.265 and While the technical merits of any video encoding method
ISO/IEC MPEG-H Part 2. In general, the marketplace will obviously inform the decision as to whether or not to
seems to have adopted H.265/HEVC as a preferred use the technology, other factors will also influence the
nomenclature. process.
Historically, it has taken roughly five years or so for a new 2.1. Patents and Licensing
video coding standard to gain traction. By many measures,
Like its predecessor, H.264, several keys parts of the
H.265/HEVC is running ahead of that, with substantial
H/265/HEVC technology are the subjected of patents.
uptake in the commercial world for distribution of video
However, again like H.264, there is a significant threshold
over networks. Amazon Prime, Netflix and the BBC all use
before any royalties are due: the first 100,000 units sold
H.265/HEVC encoding for some of their content.
each year attract no royalties2. So for all intents and
While part of the reason for the adoption of the new purposes, there seems to be little risk of a commercial
encoding format is the added support for Ultra High impact from patent costs.
Definition (UHD) TV formats such as “4K”, another
motivator is the reduced bandwidth required by lower- 2.2. Alternative Technologies
resolution formats. Clearly, if there exists a mechanism While considering the merits of H.265/HEVC, it is worth
suited to deliver the “4K” format 2160p60, which is eight also noting that there is a potential alternative sponsored by

1The Joint Collaborative Team on Video Coding (JCT-VC), 2 This refers to the largest patent pool administered by
comprising the ISO/IEC Motion Picture Expert Group (MPEG) MPEG-LA; other patents also apply, but similar considerations
and the ITU-T Video Coding Experts Group (VCEG). seem to cover these.

ETTC 2017 – European Test & Telemetry Conference

Google: the VP9 coding scheme. By contrast to the the 3.2. Transports and Containers
ISO/ITU-T methods, the intent is that VP9 is will be Both H.264/AVC and H.265/HEVC use the same
unencumbered by patent claims; whether this intention can structures to stream and store video; in this regard they are
survive a legal challenge (should one be made) remains to interchangeable, which means that audio and other
be seen. However, no assertion of infringement has been information – most notably, KLV metadata – is unaffected
made, and indeed political (anti-trust) considerations have by a change from
resulted in what appears to be a truce between the various H.264/AVC to
rights holders. H.265/HEVC. In fact, the
From a technology perspective, VP9 is designed for a design of the MPEG
different use case; while H.265/HEVC is largely driven by Transport Stream permits
the requirements of “4K” video distribution (and can be a single stream to carry
used with lower resolutions), VP9 is the inverse, being both, simultaneously, as
primarily motivated to improve the compressibility of two different programs, as
“1080p” video, but also able to be used with larger frame shown in Figure 1. (Note
sizes. Given the sponsorship by Google, it is obvious what that this makes sense in
motivates the design: Google’s YouTube.com is source of Figure 1 MPEG Transport relatively few use cases,
a huge volume of traffic, so if they can reduce that volume Stream and is included simply to
in any meaningful way, then a huge savings in total traffic indicate the possibility. Whenever it is desirable to send
is achievable. two streams of encoded video / audio / metadata, it is
However, the authors believe that, whatever the respective usually preferable to generate two distinct and independent
merits and drawbacks of H.265/HEVC versus VP9, the Transport Streams).
former will dominate based primarily on the breadth of Naturally, H.265/HEVC encoded video can simply be
support. A survey of the various implementations of each saved as an MPEG Transport Stream in a file, but it also
method show that there are simply more solutions that can be used within the “Matroska” (.MKV) and “.MP4”
encode and/or decode H.265/HEVC than there are for VP9. containers. It is not, however, supported by the “.AVI” or
Hence from a commercial standpoint, it seems reasonable “QuickTime” formats; this is mostly an indication that
to assume that, absent a specific compelling technical those containers types are largely deprecated (rather than
reason, H.265/HEVC is the preferred “successor” to any technical issue related to the newer encoding scheme)
H.264/AVC. .
This commonality with H.264/AVC means that
3. Architecture of H.265 compared with H.264 H.265/HEVC can be managed in the same way, which in
In most respects, H.265/HEVC is a direct evolution of turn means that organizations, particularly including the
H.264/AVC. While many of the individual techniques are MISB, have granted approval to use H.265/HEVC as an
refined and improved in H.265/HEVC, there is an obvious alternative to the older format[1].
correspondence between the two coding methods. 3.3. Encoding Techniques
3.1. Specifications Both H.265/HEVC and H.264/AVC use the same general
The general design specifications of H.265/HEVC and approach to encoding a video sequence: the initial frame in
H.264/AVC are outlined in Table 1. a sequence is encoded using only intra-picture methods,
and then subsequent frames use both intra-picture and
Feature H.264/AVC H.265/HEVC inter-picture compression.
Compression
Hybrid Enhanced Hybrid The most obvious difference between the two schemes is
Spatial/Temporal Spatial/Temporal that H.265/HEVC replaces 16x16 pixel macroblocks with
Model
Prediction Prediction
4K / 2160p (4,096 x 8K / 4320p (8,192 x
a new construct, called a Coding Tree Unit (CTU). CTU’s
Frame Size (Max) can be as large as 64x64 pixels, but critically within a CTU
2,304) 4,320)
“Sweet Spot”
1080p30 2160p60
the image is further divided into one or more Coding
Video Format Blocks (CBs). The size of a CB depends on the level of
Frame Rate 59.94 fps 300 fps
detail in an image. This, simply put, it allows the algorithm
Interlaced Modes Yes Reduced
50% of to use large blocks where there’s not much detail (e.g. an
Target Bandwidth 50% of H.264/AVC expanse of wall) and small blocks where there is more
MPEG-2/H.262
detail (e.g. someone’s face).
Table 1 Specifications Compared
Within each CB, additional techniques (compared to with
(Note: the ability for H.265/HEVC to code interlaced video H.264/AVC’s macroblocks) are available to encode the
streams – such as 1080i – is retained, but the image. The effectiveness of this technique is illustrated by
implementation is by selectively encoding either de- the fact that forcing an H.265/HEVC encoder to use only
interlaced frames or the individual fields, and switching 16x16 pixel CTU’s increases the output bit rate by more
between the two approaches on the fly). than 25%.
The next improvement lies with intra-frame prediction;
H.264/AVC uses eight angular prediction modes, while

ETTC 2017 – European Test & Telemetry Conference

H.265/HEVC increases the number of angular modes to expressed as a percentage of the reference (i.e.
thirty-three. H.264/AVC) rate.
Motion Vector (MV) prediction, used for inter-picture
compression, is likewise enhanced; MV’s use larger 16 bit
values, which translates to an ability to represent four times
the offset (+/- 8192 vs. +/- 2048 horizontally and +/- 512
vertically).
Clearly, all these additional variations add dramatically to
the processing power required to encode a video sequence;
depending on the sophistication of the encoder, anywhere
from 300% to 1000% increase in CPU requirements
together with significant memory accesses (in the order of
gigabytes per second) have been measured. To help offset
the real-world consequences of these increases,
H.265/HEVC includes provision for using massively Figure 2 Video Comparison Chart
parallel algorithms (and hardware), including the ability to 4.2. Results
“tile” a frame and measures to allow distinct threads to The condensed version of the BBC’s extensive analysis is
synchronize themselves along a “wavefront”. The summarized in Table 2 below.
deblocking filter has also been adapted (actually, made less
flexible) to support greater parallelization. Bitrate Reduction
Finally, a new filter, the Sample Adaptive Offset (SAO) Sequence for Equivalent Quality
filter, is applied to help reduce the appearance of common Objective BD-Rate Subjective BD-Rate
(PSNR) (MOS)
artefacts, such as banding and ringing. 2160p 47% 63%
1080p 50% 61%
4. Quality vs Bandwidth 720p 42% 56%
Image quality can be assessed in terms of an objective, 480p 39% 57%
mathematical calculation, typically the Peak Signal-to- Average 44% 59%
Noise Ratio (PSNR). However, contemporary video Table 2 Bitrates for Equivalent Quality
encoding schemes are designed to exploit the perceptual The first, and most obvious, conclusion is that the
limits of the human visual system and brain, so objective H.265/HEVC coding scheme does indeed deliver on it’s
measures can be misleading; one well known example of design goal: for a given image quality, approximately half
this phenomena is with MP3 audio compression, which is the bit rate is required compared to H.264/AVC.
deliberately “tuned” to the human ear, so it “sounds better”
The second conclusion is a little more subtle: based on the
than the PSNR measure would seem to indicate.
difference between the subjective and objective figures,
The better approach is to use a subjective analysis H.265/HEVC can be considered to be better optimized for
employing multiple real people in as controlled an the human brain than is H.264/AVC. Or, to put it another
environment as is practical, and then “average” their way, H.265/HEVC uses bits more wisely, directing greater
subjective ratings to create a useful comparison tool. detail to the visually more important areas of a video.
This sort of experiment is, of course, costly and time
consuming to manage. Fortunately, the British 5. Quality vs Channel Loss
Broadcasting Corporation (BBC) Research and
In the Test & Evaluation (T&E) community, it is
Development Department created and executed a test plan,
sometimes a challenge to ensure reliable delivery of
with the assistance of the JCT-VC, and provided the results
streaming data; many test systems use telemetry links that
in Tan et al.[2].
are prone to dropouts as the test vehicle moves through
4.1. Test Setup space. H.265/HEVC provides a couple of capabilities that
The experiments performed by the BBC used twenty help deliver quality better than H.264/AVC under such
different video sequences in a number of different conditions.
resolutions: 2160p, 1080p, 720p and 480p. For each 5.1. Error Correction
sequence, the subjects saw both the original uncompressed
The simplest approach to managing lossy communication
video, and then the compressed video at a number of
channels is to simply use some of the “saved” bandwidth
different bitrates for both H.265/HEVC and H.264/AVC.
to add sufficient error correction structures to overcome the
For each sequence and bitrate, both the PSNR and the anticipated losses. One scheme for this if the
“Mean Opinion Score” (MOS) from the subjective results communication channel is unidirectional is the forward
were calculated, and the results plotted on a chart such as error correction in IRIG 106 Chapter 7, “Packet Telemetry
the one shown in Figure 2. Downlink”; for bidirectional links, the issue typically gets
From this chart, a well-accepted model for evaluating resolved by retransmission of lost or corrupted packets.
coding efficiency, the Bjøntegaard model[3] was used to
calculate a “Bjøntegaard Delta” metric for the bit rate,

ETTC 2017 – European Test & Telemetry Conference

5.2. Subjective Consequences T&E also contends with two competing use cases: high
Perhaps more than with any other situation, channel loss is quality (for analysis of anomalies) and low bandwidth (for
measured by the impact to the consumer, i.e. the subjective telemetry).
quality, as opposed to objective measures. For instance, if The “broadcasting” approach to that situation would be to
one can compensate for a lossy channel with error acquire data using a very quality / high bitrate
correction, then, from a subjective perspective, the losses configuration, and then transcode the video after the Edit
are invisible and simply appears to have a lower bandwidth. stage to a lower quality and bitrate. Obviously, for a T&E
With subjective considerations, H.265/HEVC offers application that is suboptimal for multiple reasons,
definite advantages over H.264/HEVC. First, as the BBC including the added latency and the power and space
experiment shows, the subjective analysis (MOS) of the required for the transcoding equipment. Further, if (as is
video samples shows that comparable video quality is common) there is a desire to simultaneously record data on-
obtainable with more than the 50% bandwidth reduction board while telemetering data to the ground, another
targeted by the design. Therefore, even if only half the transcoding effort is probably appropriate, so the
bandwidth (of H.264/AVC) is allocated for a video stream, acquisition, recording and telemetry video streams would
there is additional “headroom” for managing the lossy all use different quality / bitrate settings; for example, the
channel. Alternatively, because the subjective quality of MISB “Profiles” define different “levels” for Acquisition,
the video at 50% of the H.264/AVC bitrate is “better” than Processing/Archiving and Distribution[4].
that of the H.264/AVC video, distortion introduced by lost The simplest and most effective approach for T&E
data will have a lower subjective impact than the same loss applications to resolve this situation is to produce multiple
at double the bitrate using H.264/AVC. compressed outputs from a single uncompressed input.
Secondly, the design of the H.265/HEVC Coding Tree This can be achieved with a “daisy chain” approach, but far
architecture means that, considered across a single frame, more efficient is a scheme that uses a single video
in the encoded version, more bits will be used for detail acquisition stage in front of two or more compression
(e.g. an instrument) than for background (e.g. the sky). So engines.
losing one packet’s worth of data (up to 182 bytes) is, While this design is not unique to H.265/HEVC, the
statistically, more likely to result in a loss of data in a significantly reduced additional cost in terms of power
smaller, rather than larger, piece of the image. Subjectively, consumption, etc. increases the impetuous for combined
then, data loss tends to be less visually intrusive and thus encoders, while the practicality of telemetering an full
of greater utility than would have been the case with older H.265/HEVC stream increases the likelihood of a
encoding methods. compromise in quality/bitrate that satisfies no-one.
Thirdly, because the H.265/HEVC coding uses half the A further refinement derives from the fact that, as
bandwidth compared to H.264/AVC, it is practical to use H.265/HEVC is substantially more computationally and
higher resolutions and/or better quality settings than would memory intensive to encode than H.264/AVC, a
have been the case with the earlier method. If the quality sophisticated encoder could produce both H.265/HEVC
setting is (simplistically) likened to the “fuzz” reducing the and H.264/AVC outputs from the same input. This might
clarity of a given picture, then reducing that “fuzz” or be useful if it is desirable to retain the H.264/AVC format
increasing the detail subject to “fuzz” both result in to suit an established processing workflow while adopting
increasing the ability of the human brain to resolve detail H.265/HEVC for a telemetry link. Typically, the
(for example, distinguishing between the numerals “0” and H.264/AVC would be lightly compressed and stored on the
“8”). test article, while the H.265/HEVC stream would be
encoded with the link characteristics (bandwidth, signal
6. Additional Test & Evaluation Considerations loss, etc) in mind.
Commercial broadcasting, which is undeniably the primary As noted in section 3.2 above, the outputs – the compressed
motivator for most video coding technologies, differs video streams – can be combined in a single MPEG
significantly from the CONOPS of Test and Evaluation Transport Stream, but possibly more practical would be
applications. produced and managed as two distinct network streams, so
that normal networking techniques, such as multicasting,
6.1. Video “Workflow” routing, encryption etc. can be selectively applied to the
Broadcasting tends to be oriented around a workflow various outputs.
consisting of three distinct phases: Capture, then Edit, and
6.2. High Frame Rates
finally Disseminate. Sometimes, certainly, the “Edit” phase
is performed live (e.g. with sports broadcasts), but overall As indicated in section 3.1, H.265/HEVC can support
the “consumption” of the video is separated from the frame rates up to 300fps. While that is not sufficient for
“production” of it. This separation encourages a model true high-speed video (typically 1000fps and up), it does
whereby the encoding for the ultimate consumer is distinct open up a significant new opportunity for conventional
from the acquisition phase. video systems to provide new tools for T&E applications.
Test & Evaluation applications, in contrast, tend to closely
link the producer and consumer, which implies that the
coding of video for the consumer will happen at the point
of acquisition. This would be unremarkable, except that
ETTC 2017 – European Test & Telemetry Conference
 7. Conclusion JCT-VC: Joint Collaborative Team on Video Coding
MISB: Motion Imagery Standards Board
When digital video first became a mainstream technology3, MOS: Mean Opinion Score
4-6Mbps were required to transmit a high-quality, standard MPEG: Motion Picture Experts Group
definition, MPEG2 encoded video stream. With PSNR: Peak Signal-to-Noise Ratio
H.265/HEVC, that bandwidth is more than enough for at SMPTE: Society for Motion Picture and Television Engineers
least one very good quality high definition signal, or more T&E: Test and Evaluation
signals with judicious compression settings. UHD: Ultra High Definition

H.265/HEV has absolutely delivered on its promise to use

half the bandwidth of an equivalent H.264/AVC stream,
and has introduced techniques that have improved its
subjective quality even further.
Those factors have pushed the adoption of H.265/HEVC
faster than has been predicted, and has lead to the
conclusion that the time is ripe for the adoption of the new
standard.

8. Acknowledgements
The authors acknowledges with thanks the openness of the
flight test community, in both industry and government
across the civil and military marketplace, and their
willingness to share not only the challenges of their work
but also to entertain wide-ranging potential solutions to
those challenges.
The authors would also like to thank their colleagues and
predecessors at Delta and Ampex for the innovations that
have helped make video handling in harsh environments a
solvable problem.

9. References

[1] See http://www.gwg.nga.mil/misb/faq.html#section2.2,

retrieved April 2017.
[2] Tan et al. “Video Quality Evaluation Methodology and
Verification Testing of HEVC Compression
Performance”, IEEE Transactions on Circuits and Systems
for Video Technology (TCSVT), January 2016. See
http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumbe
r=7254155. Retrieved April 2017.
[3] G. Bjøntegaard, Calculation of Average PSNR Differences
Between RD-Curves , document VCEG-M33, ITU-T SG
16/Q6, 13th VCEG Meeting, Austin, TX, USA, Apr. 2001.
[4] Motion Imagery Standards Profile, see
http://www.gwg.nga.mil/misb/misp_pubs.html, retrieved
April 2017.

10. Glossary
AVC: Advanced Video Coding
BBC: British Broadcasting Corporation
CTU: Coding Tree Unit
DOD: Department of Defense (US)
HEVC: High Efficiency Video Coding
IEC: International Electrotechnical Commission
ISO: International Standards Organization
ITU-T: International Telecommunication Union –
Telecommunication Standarization Sector

3Circa 1996, when Digital Versatile Disks (DVD) became

available. Bit rates for DVD movies are typically 4-5Mbits/sec,
with an absolute maximum of 9.8Mb/s.
ETTC 2017 – European Test & Telemetry Conference