Understanding HD

Your comprehensive guide to

High Definition on a budget

Part One
 Understanding HD with Avid 1

1
Chapter 1

Video formats
and sampling
Video formats and sampling Understanding HD with Avid 2

Perhaps one of the most baffling areas of HD

and SD, is the shorthand jargon used to
describe sampling and colour space, such as
Line 1 • • • • • • • •
RGB 4:4:4, and Y,Cr,Cb 4:2:2. Also the video Y Y Y Y Y Y Y Y
formats such as 1080/24P sound strange until CrCb CrCb CrCb CrCb

you get to know them. For a quick initiation, or Line 2 • • • • • • • •

reminder, about sampling ratios, please read Y Y Y Y Y Y Y Y
CrCb CrCb CrCb CrCb
the piece directly below.

4:2:2 sampling of luminance and colour difference signals

4:2:2 etc (Chroma sub-sampling)

The sampling rates used in digital television are described 4:1:1, used in some DV formats and DVCAM, makes Cr and
by shorthand that has, in some ways, only a tenuous Cb samples at every fourth Y sample point on every line –
connection to what it is used to describe. The numbers but still carries more chrominance detail than PAL or NTSC.
denote ratios of sampling rates, not absolute numbers,
and they need a little interpretation to understand them
all. Sometimes these ratios are referred to as ‘chrominance Line 1 Y Y Y Y Y Y Y Y Y
CrCb CrCb
(chroma) sub-sampling’.
Line 2 Y Y Y Y Y Y Y Y Y
In most instances the first number refers to luminance (Y),
CrCb CrCb
the last two refer to chrominance – the exceptions are
4:4:4, or 4:4:4:4 (more later). The first number is nearly
4:1:1 sampling
always a 4 and that means that the luminance is sampled
once for every pixel produced in the image. There are a Then another argument says that if the chrominance is sub-
very few instances where a lower sample rate is used for sampled horizontally, as in 4:1:1, why not do the same vertically
luminance. An example is HDCAM, which is generally to give a more even distribution of colour information? So
considered to use 3:1:1 sampling. Sampling at a lower rate instead of sampling both Cr and Cb on every line, they are
than the final pixel rate is known as sub-sampling sampled on alternate lines, but more frequently on each
line (at every other Y). This is 4:2:0 sampling (4:2:0 on one
The second two numbers describe the sampling
line and 4:0:2 on the next) and it is used in MPEG-2 and
frequencies of the two pure colour digitised components
most common JPEG compression schemes.
of (Red-Y) and (Blue-Y), called Cr and Cb. In line with
television’s practise of taking advantage of our eye’s
response which is more acute for luminance than for pure Line 1 Y Y Y Y Y Y Y Y Y
colour, cuts to reduce data tend to be made in the Cr Cr Cr Cr Cr
chrominance sampling rather than luminance. The most Line 2 Y Y Y Y Y Y Y Y Y
common studio sampling system is 4:2:2 where each of the Cb Cb Cb Cb Cb
two colour components is sampled coincidently with every
second luminance sample along every line. 4:2:0 provides equal colour resolution vertically
and horizontally if using square pixels
Video formats and sampling Understanding HD with Avid 3

In many cases it is very useful to have a key (or alpha) what is now the ITU-R BT.601 standard for SD sampling.
signal associated with the pictures. A key is essentially a ‘601’ defines luminance sampling at 13.5MHz (giving 720
full image but in luminance only. So then it is logical to pixels per active line) and each of the colour difference
add a fourth number 4, as in 4:2:2:4. signals at half that rate – 6.75MHz.
Technically 4:4:4 can denote full sampling of RGB or Y, Cr, The final twist in this tale is that someone then noticed
Cb component signals – but it is rarely used for the latter. that 13.5MHz was nearly the same as 14.3MHz that was 4 x
RGB may have an associated key channel, making 4:4:4:4. NTSC subcarrier. Had he looked a little further he might
have seen a much nearer relationship to 3 x PAL SC and a
Occasionally people go off-menu and do something else –
whole swathe of today’s terminology would be that much
like over-sampling which, with good processing can
different! But so it was that the number that might have
improve picture quality. In this case you might see
been 3 and should have been 1, became 4.
something like 8:8:8 mentioned. That would be making
two samples per pixel for RGB. As HD sampling rates are 5.5 times faster than those for
SD, the commonly used studio 4:2:2 sampling actually
This sampling ratio system is used for both SD and HD.
represents 74.25MHz for Y and 37.125MHz for Cr and Cb.
Even though the sampling is generally 5.5 times bigger,
4:2:2 sampling is the standard for HD studios.
Why 4? 1080I
Logic would dictate that the first number, representing a 1:1
Short for 1080 lines, interlace scan. This is the very widely
relationship with the pixels, should be 1 but, for many good
used HD line format which is defined as 1080 lines, 1920
(and some not so good) reasons, television standards are
pixels per line, interlace scan. The 1080I statement alone
steeped in legacy. Historically, in the early 1970s, the first
does not specify the frame rate which, as defined by
television signals to be digitised were coded NTSC and
SMPTE and ITU, can be 25 and 30Hz.
PAL. In both cases it was necessary to lock the sampling
frequency to that of the colour subcarrier (SC), which itself See also: Common Image Format, Interlace, ITU-R.BT 709, Table 3

has a fixed relationship to line and frame frequencies. NTSC

subcarrier is 3.579545MHz and PAL-I’s is 4.43361875MHz 1080P
and the digital systems typically sampled at 4 x NTSC SC
TV image size of 1080 lines by 1920, progressively scanned.
or 3 x PAL SC, making 14.3 and 13.3MHz respectively.
Frame rates can be as for 1080I (25 and 30Hz) as well as 24,
Then came the step to use component video Y, B-Y and R- 50, 60Hz.
Y (luminance and two pure colour components – known as
See also: Common Image Format, Progressive, ITU-R.BT 709, Table 3
colour difference signals) that is much easier to process for
re-sizing, smooth positioning, standards conversion,
compression and all the other 1001 operations that can be
13.5MHz
applied to pictures today. When a standard was developed Sampling frequency used in the 601 digital coding of SD
for sampling this component video it followed some of the video. The frequency was chosen to be a whole multiple of
same logic as before, but this time also sort commonality the 525 and 625-line television system frequencies to
between the two SD scanning systems used around the create some compatibility between the digital systems.
world: 525/60I and 625/50I. Putting all that together led to The sampling is fast enough to faithfully portray the
Video formats and sampling Understanding HD with Avid 4

highest frequency, 5.5MHz, luminance detail information The images are a pure electronic equivalent of a film shoot
present in SD images. Digital sampling of most HD and telecine transfer – except the video recorder operates
standards samples luminance at 74.25MHz, which is 5.5 at film rate (24 fps), not at television rates. The footage has
times 13.5MHz. more of a filmic look but with the low frame rate,
movement portrayal can be poor.
See also: 2.25MHz, ITU-R BT.601
25PsF and 30PsF rates are also included in the ITU-R BT.
2.25MHz 709-4 recommendation.
See also: ITU-R BT. 709
This is the lowest common multiple of the 525/59.94 and
625/50 television line frequencies, being 15.734265kHz and
15.625kHz respectively. Although seldom mentioned, its
601
importance is great as it is the basis for all digital See ITU-R BT. 601
component sampling frequencies both at SD and HD.
See also: 13.5MHz 709
See ITU-R BT. 709
24P
Short for 24 frames, progressive scan. In most cases this 720P
refers to the HD picture format with 1080 lines and 1920
Short for 720 lines, progressive scan. Defined in SMPTE
pixels per line (1080 x 1920/24P). The frame rate is also
296M and a part of both ATSC and DVB television standards,
used for SD at 480 and 576 lines with 720 pixels per line.
the full format is 1280 pixels per line, 720 lines and 60
This is often as an offline for an HD 24P edit, or to create a
progressively scanned pictures per second. It is mainly the
pan-and-scan version of an HD down-conversion. Displays
particular broadcasters who transmit 720P that use it. Its 60
working at 24P usually use the double shuttering technique
progressive scanned pictures per second offers the benefits
– like film projectors – to show each image twice and
of progressive scan at a high enough picture refresh rate to
reduce flicker when viewing this low rate of images.
portray action well. It has advantages for sporting events,
smoother slow motion replays etc.
24PsF
24P Segmented Frame. This blurs some of the boundaries 74.25MHz
between film/video as video is captured in a film-like way,
The sampling frequency commonly used for luminance (Y)
formatted for digital recording and can pass through
or RGB values of HD video. Being 33 x 2.25MHz, the
existing HD video infrastructure. Like film, entire images
frequency is a part of the hierarchical structure used for SD
are captured at one instant rather than by the usual line-
and HD. It is a part of SMPTE 274M and ITU-R BT.709.
by-line TV scans down the image that means the bottom
can be scanned 1/24 of a second after the top. The images See also: 2.25MHz

are then recorded to tape as two temporally coherent

fields (segments), one with odd lines and the other with
even lines, that are well suited to TV recorders.
Video formats and sampling Understanding HD with Avid 5

Active picture Anamorphic

The part of the picture that contains the image. With the This generally describes cases where vertical and
analogue 625 and 525-line systems only 575 and 487 lines horizontal magnification is not equal. The mechanical
actually contain the picture. Similarly, the total time per anamorphic process uses an additional lens to compress
line is 64 and 63.5µS but only around 52 and 53.3µS contain the image by some added amount, often on the horizontal
picture information. As the signal is continuous the extra axis. In this way a 1.85:1 or a 2.35:1 aspect ratio can be
time allows for picture scans to reset to the top of the squeezed horizontally into a 1.33:1 (4:3) aspect film frame.
frame and the beginning of the line. When the anamorphic film is projected it passes through
another anamorphic lens to stretch the image back to the
Digitally sampled SD formats contain 576 lines and 720
wider aspect ratio. This is often used with SD widescreen
pixels per line (625-line system), and 480 lines and 720 pixels
images which keep to the normal 720 pixel count but
per line (525-line system) but only 702 contain picture
stretch them over a 33-percent wider display. It can also
information. The 720 pixels are equivalent to 53.3µS.
apply to camera lenses used to shoot 16:9 widescreen
The sampling process begins during line blanking of the where the CCD chips are 4:3 aspect ratio.
analogue signal, just before the left edge of active picture,
See also: Aspect ratio
and ends after the active analogue picture returns to
blanking level. Thus, the digitised image includes the left
and right frame boundaries as part of the digital scan line. Anti-aliasing
This allows a gentle roll-on and roll-off the between the
Attempts to reduce the visible effects of aliasing. This is
blanking (black) and active picture.
particularly the case with spatial anti-aliasing that typically
HD systems are usually quoted just by their active line uses filtering processes to smooth the effects of aliasing
count, so a 1080-line system has 1080 lines of active video, which may be noticeable as jaggedness on diagonal lines,
each of 1920 samples. This may be mapped onto a larger or ‘twinkling’ on areas of fine detail. A better solution is to
frame, such as 1125 lines, to fit with analogue connections. improve the original sampling and processing and avoid
aliasing in the first place.
Aliasing See also: Aliasing

Artefacts created as a result of inadequate or poor video

sampling or processing. Spatial aliasing results from the Aspect Ratio
pixel-based nature of digital images and leads to the
For pictures, this refers to the ratio of picture width to
classic ‘jagged edge’ (a.k.a. ‘jaggies’) appearance of
height. HD pictures use a 16:9 aspect ratio, which also may
curved and diagonal detail and twinkling on detail. This
be noted as 1.77:1. This is a third wider than the traditional
results from sampling rates or processing accuracy too low
4:3 television aspect ratio (1.33:1) and is claimed to
for the detail. Temporal aliasing occurs where the speed of
enhance the viewing experience as it retains more of our
the action is too fast for the frame rate, the classic
concentration by offering a wider field of view.
example being wagon wheels that appear to rotate the
wrong way. Pixel aspect ratio refers to the length versus height for a
pixel in an image. HD always uses square pixels as do
See also: Anti-aliasing
most computer applications. SD does not. The matter is
Video formats and sampling Understanding HD with Avid 6

further complicated by SD using 4:3 and 16:9 (widescreen) maintain the dynamic range. For example, if the YCrCb
images which all use the same pixel and line counts. Care colour space video is 8 bits per component then the RGB
is needed to alter pixel aspect ratio when moving between colour space video will need to be 10 bits.
systems using different pixel aspect ratios so that objects
retain their correct shape.
Component video
With both 4:3 and 16:9 images and displays in use, some
Most traditional digital television equipment handles video
thought is needed to ensure a shoot will suit its target
in the component form: as a combination of pure luminance
displays. All HD, and an increasing proportion of SD,
Y, and the pure colour information carried in the two colour
shoots are 16:9 but many SD displays are 4:3. As most HD
difference signals R-Y and B-Y (analogue) or Cr, Cb (digital).
productions will also be viewed on SD, clearly keeping the
The components are derived from the RGB delivered by
main action in the middle ‘4:3’ safe area would be a good
imaging devices, cameras, telecines, computers etc.
idea – unless the display is letterboxed.
Part of the reasoning for using components is that it allows
See also: ARC
colour pictures to be compressed. The human eye can see
much more detail in luminance than in the colour information
Chrominance (or Chroma) (chrominance). The simple task of converting RGB to Y, (R-

sub-sampling Y) and (B-Y) allows exclusive access to the chrominance

only, so its bandwidth can be reduced with negligible impact
See 4:2:2 etc. on the viewed pictures. This is used in PAL and NTSC
colour coding systems and has been carried through to
CIF component digital signals both at SD and HD.

Common Image Format. An image format that is widely For the professional digital video applications, the colour
used and denoted ‘Common Image Format’ by the ITU. difference signals are usually sampled at half the frequency
The idea is to promote the easy exchange of image of the luminance - as in 4:2:2 sampling. There are also other
information nationally and internationally. types of component digital sampling such as 4:1:1 with less
colour detail (used in DV), and 4:2:0 used in MPEG-2.
See HD-CIF

Co-sited sampling
Colour space
Where samples of luminance and chrominance are all
The space encompassed by a colour system. Examples taken at the same instant. This is designed so that the
are: RGB, YCrCb, HSL (hue, saturation and luminance) for relative timing (phase) of all signal components is
video, CMYK for print and XYZ for film. Moving between symmetrical and not skewed by the sampling system.
media, platforms or applications can require a change of Sampling is usually co-sited but there is a case of 4:2:0
colour space. This involves complex image processing so sampling being interstitial – with chrominance samples
care is needed to get the right result. Also, repeated made between the luminance samples.
changes of colour space can lead to colours drifting off.
See also: 4:2:2 etc.
It is important to note that when converting from YCrCb to
RGB more bits are required in the RGB colour space to
Video formats and sampling Understanding HD with Avid 7

DTV As HD’s 1080 x 1920 image size is close to the 2K used for
film, there is a crossover between film and television. This
Digital Television. This is a general term that covers both is even more the case if using a 16:9 window of 2K as here
SD and HD digital formats. there is very little difference in size. It is generally agreed
that any format containing at least twice the standard
Gamut (colour) definition format on both H and V axes is high definition.

The range of possible colours available in an imaging After some initial debate about the formats available to
system. The red, blue and green phosphors on television prospective HD producers and television stations, the
screens and the RGB colour pick-up CCDs or CMOS chips acceptance of 1080-HD video at various frame rates, as a
in cameras, define the limits of the colours that can be common image format by the ITU, has made matters far
displayed – the colour gamut. Between the camera and more straightforward. While television stations may have
viewer’s screen there are many processes, many using some latitude in their choice of format, translating, if
component 4:2:2 video. However, not all component value required, from the common image formats should be
combinations relate to valid RGB colours (for example, routine and give high quality results.
combinations where Y is zero). Equipment that generates
2048
images directly in component colour space, such as some
graphics machines, can produce colours within the 2K Film
component range but that are invalid in RGB, which can
also exceed the limits allowed for PAL and NTSC.
1920
There is potential for overloading equipment – especially
transmitters which may cut out to avoid damage! There is 1080-HD
equipment that clearly shows many areas of out-of-gamut

1536
pictures, so that they can be adjusted before they cause 1280

problems. 720 720-HD

1080
576
HD &

720
480-SD

576
480
High Definition Television. This has been defined in the
USA by the ATSC and others as having a resolution of
approximately twice that of conventional television (meaning
analogue NTSC – implying 486 visible lines) both horizontally 2K, HD and SD images sizes

and vertically, a picture aspect ratio of 16:9 and a frame rate

See also: Common Image Format, Interlace Factor
of 24fps and higher. This is not quite straightforward as the
720-line x 1280 pixels per line, progressive scan format is
well accepted as HD. This is partly explained by the better
vertical resolution of its progressive scanning. Apart from
the video format, another HD variation on SD is a slightly
different colorimetry where, for once the world agrees on a
common standard.
Video formats and sampling Understanding HD with Avid 8

PAL and NTSC RGB

PAL and NTSC do not exist in HD. They do not exist in Red, Green and Blue. Cameras, telecines and most
modern SD digital television either – although it was computer equipment originate images in this colour
digitised in early digital VTR formats. PAL, means Phase space. For digital sampling, all three colours are sampled
Alternating Line and is an analogue system for coding in the same way at full bandwidth – hence 4:4:4. images
colour that is still widely in use. Similarly NTSC (National may offer better source material for the most critical
Television Standards Committee) describes an analogue chroma keying, but they occupy 50 percent more data
system. Confusingly PAL and NTSC are still used to space than 4:2:2 and as no VTRs record 4:4:4, data
describe frame rates and formats that relate in some way recorders or disks must be used to store them. Also, there
with their analogue world. So 1080 PAL might be 1080/50I. are no television means to connect them, so IT-based
networking technology is used.

Quantization Often 4:4:4 is only used in post production areas and is

converted to 4:2:2 when material is more widely distributed.
Quantization refers to sampling: the number of bits used in
See also: 4:4:4, Gamut
making digital samples of a signal. For video, 8 bits is
quite common in consumer and prosumer products such
as DV. HDV also uses 8 bits. Note that the 8 bits can define
28 or 256 numbers or levels that, for converting analogue
Segmented Frame
video into digits, are assigned to levels of image brightness. See: 24PsF

For more accuracy and to withstand multiple levels of complex

post production processing, studio video applications Square pixels
often use 10-bit sampling – providing 1024 levels.
Square pixels are the pixel aspect ratio where the pixels
Usually the distribution of the levels between brightest and describe a square area of the displayed image. This is the
darkest is linear (even) but in the case of scanning film case with HD broadcast standards, as the picture formats
negative for input to a digital intermediate chain, then a describe line length (number of pixels per line) and
logarithmic distribution is often used that progressively number of lines, in exact 16:9 ratios – which is also the
squashes the levels into the darker areas of picture. This is display aspect ratio of the pictures.
because film negative has to carry a very wide range of
There are places in HD where pixels are not square. The
contrast information from the original scene, and the levels
very widely used HDCAM sub-samples the 1920-pixel HD
in the dark/shadow areas are more significant and visible
line lengths with 1440 luminance samples. This is only an
than those in bright areas. The ‘log’ sampling suitably
internal function of the recorder; the inputs and outputs
redistributes the available digital levels – hence 10-bit log.
use square pixels. In a similar way the 1080I HDV(2) format
This is considered to be as useful as 13-bit linear quantization.
also uses 1440 samples per line.
NB: Quantization has another meaning.
Generally, computers generate images with square pixels
See section: Video Compression 1
but digital SD television images are not square. This
means that any applications or equipment used needs to
Video formats and sampling Understanding HD with Avid 9

take this into account when transferring between applications,

or performing image manipulations to maintain correct
Table 3
image aspect ratios (so circles remain circular). The video formats allowed for broadcast in the ATSC DTV
standard are listed in Table 3 of document Doc. A/53A.
See also: Anamorphic, Aspect ratio
Table 3 Compression Format Constraints
Vertical_ Horizontal_ aspect_ frame_ progressive_
Sub-sampling size_
value
size_
value
ratio_
information
rate_
code
sequence

In a digital sampling system, taking fewer samples of an

1080 1920 1,3 1,2,4,5 1
analogue signal than the number of pixels in the digital 4,5 0
image is called sub-sampling. Generally sub-sampling is 720 1280 1,3 1,2,4,5,7,8 1
used to reduce the amount of data used for an image. In 1,2,4,5,7,8 1
the widely used 4:2:2 sampling system for studio quality 480 704 2,3 4,5 0
video, each luminance sample corresponds to one pixel – 640 1,2 1,2,4,5,7,8 1
denoted by the ‘4’. The two chrominance signals are each 4,5 0
sampled at half the rate, making one per two pixels. This is
known as chrominance sub-sampling – a term that is
sometimes more generally ascribed to the sampling ratios Legend for MPEG-2 coded values in Table 3
– such as 4:2:2, 4:1:1, etc. aspect_ratio_information 1 = square samples 2 = 4:3 display aspect ratio 3 = 16:9 display aspect ratio

Frame_rate_code 1 = 23.976 Hz 2 = 24 Hz 4 = 29.97 Hz 5 = 30 Hz 7 = 59.94 Hz 8 = 60 Hz

See also: 4:2:2 etc
Progressive_sequence 0 = interlaced scan 1 = progressive scan

System nomenclature This table lists no fewer than 18 DTV formats for SD and HD.
Initially, this led to some confusion about which should be
A term used to describe television standards. The
adopted for whatever circumstances. Now most HD
standards are mostly written in a self-explanatory form but
production and operation is centred on the 1080-line formats
there is room for confusion concerning vertical scanning
either with 24P, 25P or 60I vertical scanning, and 720-line
rates. For example, 1080/60I implies there are 60 interlaced
formats at 50P and 60P.
fields per second that make up 30 frames. Then 1080/30P
describes 30 frames per second, progressively scanned.
The general rule appears to be that the final figure always
indicates the number of vertical refreshes per second.
However, Table 3 (below) uses a different method. It defines
frame rates (numbers of complete frames) and then defines
whether they are interlaced or progressive. So here the ‘frame
rate code 5’ is 30Hz which produces 30 vertical refreshes
when progressive, and 60 when interlaced. Be careful!
See also: Interlace, Progressive
Video formats and sampling Understanding HD with Avid 10

Truncation (a.k.a. Rounding) Universal Format

Reducing the number of bits used to describe a value. This is 1080/24P is sometimes referred to as the Universal Format for
everyday practice; we may say 1,000 instead of 1024 in the television. The reason is its suitability for translation into all
same way we leave off the cents/pence when talking about other formats to produce high quality results in all cases.
money. There is also the need to truncate the digits used in
See also: HD-CIF, Universal Master
digital video systems. With due care, this can be invisible,
without it degradation becomes visible.
Decimal: 186 x 203 = 37758 Universal Master
Binary: 10111010 x 11001011 = 1001001101111110 The 1080/24P format has well defined and efficient paths to all
It is the nature of binary mathematics that multiplication, which major television formats and is capable of delivering high
is commonplace in video processing (e.g. mixing pictures), quality results to all. An edited master tape in this format is
produces words of a length equal to the sum of the two sometimes referred to as a Universal Master.
numbers. For instance, multiplying two 8-bit video values See also: HD-CIF
produces a 16-bit result – which will grow again if another
process is applied. Although hiways within equipment may
carry this, ultimately the result will have to be truncated to fit Y, Cr, Cb
the outside world which, for HD, may be a 10-bit HD-SDI
This signifies video components in the digital form. Y, Cr, Cb is
interface or 8-bit MPEG-2 encoder.
the digitised form of Y, R-Y, B-Y.
In the example, truncating by dropping the lower eight bits
lowers its value by 01111110, or 126. Depending on video
content, and any onward processing where the error is Y, R-Y, B-Y
compounded, this may, or may not be visible. Typically, flat (no See component video
detail) areas of low brightness are prone to showing this type
of discrepancy as banding. This is, for example, sometimes
visible from computer generated images. YUV
Inside equipment, it is a matter of design quality to truncate De-facto shorthand for any standard involving component
numbers in an intelligent way that will not produce visible video. This has been frequently, and incorrectly, used as
errors – even after further processing. Outside, plugging 10-bit shorthand for SD analogue component video – Y, R-Y, B-Y. Y is
equipment into 8-bit needs care. Intelligent truncation is correct, but U and V are axes of the PAL colour subcarrier
referred to as Rounding. which are modulated by scaled and filtered versions of B-Y
and R-Y respectively. Strangely, the term is still used to
describe component analogue HD. This is double folly.
Although Y is still correct, all HD coding is digital and has
nothing to do with subcarriers or their axes. So forget it!
 Understanding HD with Avid 11

2
Chapter 2

Video Compression:
Concepts
Video Compression: Concepts Understanding HD with Avid 12

Video compression reduces the amount of data the digital video compression schemes in use today
or bandwidth used to describe moving pictures. including AVR, DV, HDV, JPEG (but not JPEG2000) and the
I frames of MPEG-1, 2 and 4, and Windows Media 9. A
Digital video needs vast amounts of data to
further reduction is made using Huffman coding, a purely
describe it and there have long been various
mathematical process that reduces repeated data.
methods used to reduce this for SD. And as HD
MPEG-2 and the more recent MPEG-4 add another layer
has up to a six times bigger requirement of
of compression by analysing what changes form frame to
1.2Gb/s and requiring 560GB per hour of storage,
frame by analysing the movement of 16 x 16-pixel macro
the need for compression is even more pressing. blocks of the pictures. Then it can send just the movement
information, called motion vectors, that make up predictive
(B and P) frames and contain much less data than I frames,
Intro Compression – General for much of the time. Whole pictures (I frames, more data)
are sent only a few times a second. MPEG-2 compression
Exactly which type and how much compression is used
is used in all forms of digital transmission and DVDs as well
depends on the application. Consumer delivery (DVD,
as for HDV. The more refined and efficient MPEG-4 is now
transmission, etc) typically uses very high compression (low
being introduced for some HD services, and is set to
data rates) as the bandwidth of the channels is quite small.
become widely used for new television services.
For production and online editing use much lighter
compression (higher data rates) are used as good picture Each of these techniques does a useful job but needs to
quality needs to be maintained though all the stages be applied with some care when used in the production
leading to the final edited master. chain. Multiple compression (compress/de-compress)
cycles may occur while moving along the chain, causing a
Video compression methods are all based on the principle
build-up of compression errors. Also, as many compression
of removing information that we are least likely to miss –
schemes are designed around what looks good to us, they
so-called ‘redundant’ picture detail. This applied to still
may not be so good in production, post production and
images as well as video and cinema footage. This takes the
editing. This particularly applies in processes, such as
form of several techniques that may be used together.
keying and colour correction, that depend on greater
Digital technology has allowed the use of very complex
image fidelity than we can see, so disappointing results
methods which have been built into low cost mass
may ensue from otherwise good-looking compressed
produced chips.
originals.
First, our perception of colour (chroma) is not as sharp as it
See also: AVR, Component video, DV, DNxHD, Huffman coding, JPEG,
is for black and white (luminance), so the colour resolution
JPEG2000, MPEG-2, MPEG-4
is reduced to half that of luminance (as in 4:2:2). This is used
in colour television (NTSC, PAL and digital). Similarly, fine
detail with little contrast is less noticeable than bigger
objects with higher contrast. To access these a process
Blocks
called DCT resolves 8 x 8 pixel blocks of digital images into See DCT

frequencies and amplitudes to make it possible to scale

(down), or ‘quantize’, the DCT coefficients (frequencies and
amplitudes) and so reduce the data. This applies most of
Video Compression: Concepts Understanding HD with Avid 13

Codec Compression-friendly
Codec is short for coder/decoder – usually referring to a Material that looks good after compression is sometimes
compression engine. Confusingly, the term is often misused referred to as ‘compression friendly’. This can become
to describe just a coder or decoder. important in transmission where very limited data
bandwidth is available and high compression ratios have to
be used. Footage with large areas of flat colour, little detail
Compression ratio and little movement compress very well: for example,
This is the ratio of the uncompressed (video or audio) data cartoons, head-and-shoulder close-ups and some dramas.
to the compressed data. It does not define the resulting As, MPEG-2 compression looks at spatial detail as well as
picture or sound quality, as the effectiveness of the movement in pictures and an excess of both may show at
compression system needs to be taken into account. Even the output as poor picture quality. This often applies to
so, if used in studio applications, compression is usually fast-moving sports – for instance football.
between 2:1 and 7:1 for SD (and D1 and D5 uncompressed Poor technical quality can be compression unfriendly.
VTRs are also available), whereas compression for HD is Random noise will be interpreted as movement by an
currently approximately between 6:1 and 14:1 – as defined MPEG-2 or MPEG-4 encoder, so it wastes valuable data
by VTR formats, and is I-frame only. For transmission, the space conveying unwanted movement information.
actual values depend on the broadcaster’s use of the Movement portrayal can also be upset by poor quality
available bandwidth but around 40:1 is common for SD frame-rate conversions that produce judder on movement,
and somewhat higher, 50 or 60:1 for HD (also depending again increasing unwanted movement data to be
on format). These use both I-frames and the predictive transmitted at the expense of spatial detail. Such
frames to give the greater compression. circumstances also increase the chance of movement
HDV records data to tape at 19-25 Mb/s – a rate going wrong – producing ‘blocking’ in the pictures.
comparable with HD transmission and a compression ratio Errors can be avoided by the use of good quality
of around 40:1, depending on the standard used. equipment throughout the production chain. Also, the
Transmission and video recorders in general work at a choice of video format can help. For example, there is less
constant bit rate so, as the original pictures may include movement in using 25 progressively scanned images than
varying amounts of detail, the quality of the compressed in 50 interlaced fields, so the former compress more easily.
images varies. DVDs usually work on a constant The efficiency increase is typically 15-20 percent.
quality/variable bit rate principle. So the compression ratio
slides up and down according to the demands of the
material, to give consistent results. This is part of the
DCT
reason why DVDs can look so good while only averaging Discrete Cosine Transform is used as a first stage of many
quite low bit rates – around 4 Mb/s. digital video compression schemes including JPEG and
MPEG-2 and –4. It converts 8 x 8 pixel blocks of pictures to
express them as frequencies and amplitudes. This may not
reduce the data but it does arrange the image information
so that it can. As the high frequency, low amplitude detail
is least noticeable their coefficients are progressively
Video Compression: Concepts Understanding HD with Avid 14

reduced, some often to zero, to fit the required file size Studio applications of MPEG-2 have very short GOPs,
per picture (constant bit rate) or to achieve a specified Betacam SX has a GOP of 2, IMX has 1, (i.e. I-frame only –
quality level. It is this reduction process, known as no predictive frames) which means cutting at any frame is
quantization, which actually reduces the data. straightforward. Other formats such as DV, DVCPRO HD
and HDCAM, D5-HD do not use MPEG but are also I-
For VTR applications the file size is fixed and the
frame only.
compression scheme’s efficiency is shown in its ability to
use all the file space without overflowing it. This is one See also: MPEG-2, MPEG-4
reason why a quoted compression ratio is not a complete
measure of picture quality.
DCT takes place within a single picture and so is intra-
I-frame only (aka I-frame)
frame (I-frame) compression. It is a part of the currently Short for intra-frame only.
most widely used compression in television.
See also: AVR, Compression ratio, DV, JPEG, MPEG-2, MPEG-4 Inter-frame compression
Video compression that uses information from several
GOP successive video frames to make up the data for its
compressed ‘predictive’ frames. The most common
Group Of Pictures – as in MPEG-2 and MPEG-4 video
example is MPEG-2 with a GOP greater than 1. Such an
compression. This is the number of frames to each integral
MPEG-2 stream contains a mix of both I-frames and
I-frame: the frames between being predictive (types B and
predictive B and P (Bi-directional predictive and Predictive)
P). ‘Long GOP’ usually refers to MPEG-2 and 4 coding. For
frames. Predictive frames cannot be decoded in isolation
transmission the GOP is often as long as half a second, 13,
from those in the rest of the GOP so the whole GOP must
or 15 frames (25 or 30fps), which helps to achieve the
be decoded. This is an efficient coding system that is good
required very high compression ratios.
for transmission but it does not offer the flexibility needed
for accurate editing as it can only be cut at the GOP
boundaries. It also requires estimation of the movement
from picture to picture, which is complex and not always
I B B P B B P B B P B B I accurate – leading to ‘blockiness’.
See also: GOP, MPEG-2, MPEG-4

A typical group of pictures

Cutting long GOP MPEG is not straightforward as its

Interlace
accuracy is limited to the GOP length unless further A method of ordering the lines of scanned images as two
processing is applied – typically decoding. HDV uses long (or more) interlaced fields per frame. Most television uses
GOP MPEG-2 of 6 or 15 frames for HDV1 or HDV2 2:1 interlacing; alternate fields of odd lines 1,3,5, etc.,
respectively making it editable at 1/4 or 1/2 second followed by a field of even lines 2, 4, 6, etc. This doubles
intervals. A GOP of 1 indicates ‘I-frame only’ video, which the vertical refresh rate as there are twice as many
can be cut at every frame without need of processing. interlaced fields as there would be whole frames. The
Video Compression: Concepts Understanding HD with Avid 15

result is better portrayal of movement and reduction of

flicker without increasing the number of full frames or
Macroblock
required signal bandwidth. There is an impact on vertical A 16 x 16 pixel block, comprising four adjacent DCT blocks
resolution and care is needed in image processing. – macroblocks are used to generate motion vectors in
MPEG-2 coding. Most coders use a ‘block matching’
See also: Interlace factor, Progressive
technique to establish where the block has moved and so
generate motion vectors to describe the movement. This
Interlace Factor works most of the time but also has its well-known
moments of failure. For example, slow fades to black tend
Use of interlaced, rather than progressive, scans has no to defeat the technique, making the resulting misplaced
effect on the vertical resolution of still images. However, if blocks quite visible. Better technologies are available for
anything in the image moves the resolution is reduced by use in movement estimation, such as phase correlation.
the Interlace Factor, which may be 0.7 or less. This is due
to the time displacement between the two fields of
interlace which will produce detail that is jagged, line-by- Motion Vectors
line, during the movement and it appears as an overall
Used in MPEG-2 and MPEG-4 compression systems, motion
slight softening of vertical resolution.
vectors describe the direction and distance that macroblocks
(16 x 16 pixels) move between frames. Sending this
movement information requires much less data than sending
Intra-frame compression an I frame, and so further reduces the video data.
(a.k.a. I-frame compression)
Video compression which takes information from one
video frame only. This way, all the information to re-create
Progressive (scan)
the frame is contained within its own compressed data and Sequence for scanning an image where the vertical scan
is not dependent on other adjacent frames. This means progresses from line 1 to the end in one sweep. In HDTV
that I-frame compressed video is easily edited as it can there are a number of progressive vertical frame (refresh)
simply be cut at any picture boundary without the need for rates allowed and used. 24Hz is popular for its
any decoding and recoding. I-frame only video can be compatibility with motion pictures and its ability to be
edited and the result output as first generation material. easily translated into all of the world’s television formats.
Any other operations such as wipes, dissolves, mixes, DVE 25 and 30Hz correspond with existing SD frame rates
moves etc., can only be performed on the baseband (although they use interlaced scans). 50 and 60Hz are also
signal, requiring that the video is first decompressed. allowed for, but, due to bandwidth restrictions, these are
limited in picture size, e.g. 720/60P and 720/50P.
See also: AVR, DV, JPEG, MPEG-2,
Today, progressive scanning is most commonly found in
computer displays and all the modern panel TV displays are
progressive. Progressive images are rock steady making the
detail easy to see. For the equipment designer progressive
images are easier process as there is no difference between
the two fields of a frame to contend with.
Video Compression: Concepts Understanding HD with Avid 16

Progressive scans have the disadvantage of a slow vertical

refresh rate. Thus, for the lower rates of 24, 25 and 30Hz,
which can be used in HD television with the larger 1080-
line formats, there would be considerable flicker on
displays, unless there were some processing to show each
picture twice (as in double shuttering in cinema projectors).
Besides flicker, the other potential problem area is that of
fast action or pans, as the lower refresh rate means that
movement will tend to stutter. Interlace, with its two
vertical refreshes per frame, has advantages here.
See also: 24PsF, Interlace

Quantization
Quantizing is the process used in DCT-based compression
schemes, including AVC, JPEG, MPEG-2 and MPEG-4, to
reduce the video data in an I frame. DCT allows quantizing
to selectively reduce the DCT coefficients that represent
the highest frequencies and lowest amplitudes that make
up the least noticeable elements of the image. As many
are reduced to zero significant data reduction is realised.
Using a fixed quantizing level will produce a constant
quality of output with a data rate that varies according to
the amount of detail in the images. Alternatively
quantizing can be varied to produce a constant data rate,
but variable quality, images. This is useful where the data
must be fitted into a defined size of store or data channel
– such as a VTR or a transmission channel. The success in
nearly filling, but never overflowing, the storage is one
measure of the efficiency of DCT compression schemes.
NB: Quantization has a second meaning.
See Video Formats section
 Understanding HD with Avid 17

3
Chapter 3

Video Compression:
Formats
Video Compression: Formats Understanding HD with Avid 18

DVC
DVC is the compression used in DV equipment that is
standardised in IEC 61834. It is a DCT-based, intra-frame
scheme achieving 5:1 compression so that 8-bit video
This is the practical side of compression showing sampling of 720 x 480 at 4:1:1 (NTSC) or 720 x 576 4:2:0
the systems and formats that are used. Some are (PAL) produces a 25 Mb/s video data rate. The same is
proprietary, in which case the company involved used for DV, DVCAM, Digital8 and DVCPRO (where PAL is
is mentioned. PAL 4:1:1). It achieves good compression efficiency by
applying several quantizers at the same time, selecting the
nearest result below 25Mb/s for recording to tape.

AVC
See MPEG-4 DNxHD
Avid DNxHD encoding is designed to offer quality at
significantly reduced data rate and file size and it is
AVR supported by the family of Avid editing systems.
AVR is a range of Motion-JPEG video compression Engineered for editing, it allows any HD material to be
schemes devised by Avid Technology for use in its ABVB handled on SD-original Avid systems. Any HD format can
hardware-based non-linear systems. An AVR is referred to be encoded edited, effects added, colour corrected and
as a constant quality M-JPEG resolution since the same the project finished.
quantization table (of coefficients) is applied to each frame
There is a choice of compression image formats to suit
of a video clip during digitization. For any given AVR, the
requirements. Some of the formats are:
actual compressed data rate will increase as the
complexity of the imagery increases. For example, a head
Format DNxHD DNxHD DNxHD DNxHD DNxHD
shot typically results in a low data rate while a crowd shot 220x 185x 185 145 120
from a sporting event will yield a high data rate. To avoid Bit depth 10 bit 10 bit 8 bit 8 bit 8 bit
system bandwidth problems, AVRs utilize a mode of rate Frame rate 29.92 fps 25 fps 25 fps 25 fps 25 fps
control called rollback which prevents the compressed Data rate 220 Mb/s 184 Mb/s 184 Mb/s 135 Mb/s 220 Mb/s
data rate from increasing beyond a preset limit for a
sustained period. So, when the data rate exceeds the Avid DNxHD maintains the full raster, is sampled at 4:2:2
rollback limit on a given frame, high spatial frequency and uses highly optimised coding and decoding
information is simply discarded from subsequent frames techniques, so image quality is maintained over multiple
until the rate returns to a tolerable level. generations and processes. When you’re ready, master to
See also: DCT, JPEG any format you need.
DNxHD efficiency enables collaborative HD workflow using
networks and storage designed to handle SD media. So,
for example, Avid Unity shared media networks are HD-
ready today! Cost-effective, real-time HD workflows can be
Video Compression: Formats Understanding HD with Avid 19

designed with Media Composer Adrenaline HD and Avid

DS Nitris systems. You can even edit HD on a laptop.
JPEG
Joint (ISO and ITU-T) Photographic Experts Group; JPEG is
For more information see www.avid.com/dnxhd/index.asp
a standard for compressing still picture data. It offers
compression ratios of between two and 100 times and
H.264 there are three levels of processing available: baseline,
extended and lossless encoding.
See MPEG-4
JPEG baseline coding, which is the most common for
television and computer applications, starts by applying
Huffman coding DCT to 8 x 8 pixel blocks of the picture, transforming them
into frequency and amplitude data. This itself may not
A method of compressing data by recognizing repeated
reduce data but then the generally less visible high
patterns and assigning short codes to those that occur
frequencies can be divided by a high quantizing factor
frequently, and longer codes to those that are less
(reducing many to zero), and the more visible low
frequent. The codes are assigned according to a Huffman
frequencies by a lower factor. The quantizing factor can be
Table. Sending the codes rather than all the original data
set according to data size (for constant bit rate) or picture
can achieve as much as a 2:1 lossless compression and the
quality (constant quality) requirements – effectively
method is often used as a part of video compression
adjusting the compression ratio. The final stage is Huffman
schemes such as JPEG and MPEG.
coding which is a lossless mathematical treatment that can
further reduce data by 2:1 or more.
JFIF Baseline JPEG coding creates .jpg files and is very similar
to the I-frames of MPEG-1, -2 and -4, the main difference
JPEG File Interchange Format – a compression scheme
being they use slightly different Huffman tables.
used by Avid Technology in its Meridien hardware-based
non-linear systems. A JFIF M-JPEG resolution is termed See also: Compression, Compression ratio, DCT, DV, Huffman coding,
constant rate since compressing clips of varying JFIF, M-JPEG,
complexity results in a fixed data rate. Each JFIF resolution
is defined by a target data rate and a base quantization
table. When digitizing, the quantization table is linearly www http://www.jpeg.org

scaled (known as rolling Q) to conform the actual

compressed data rate to the target rate. Due to the
flexibility of this approach, imagery compressed by a JFIF
JPEG 2000
resolution generally looks better than that compressed by JPEG 2000 is an advanced image coding (compression)
an AVR of comparable average data rate. system from the Joint Photographic Experts Group. Like
‘normal’ JPEG, this is intra-frame compression and it is
suitable for a wide range of uses from portable digital
cameras, to scientific and industrial applications.
Video Compression: Formats Understanding HD with Avid 20

Rather than using the established DCT, it employs state-of-

the-art techniques based on wavelet technology. Requiring
MPEG
more processing than MPEG, JPEG 2000 has, until recently Moving Pictures Expert Group. A group of industry experts
been too costly for wide use in television applications. involved with setting standards for moving pictures and
Now new chips have lowered the price barriers and JPEG sound. These are not only those for the compression of
2000’s use in TV and D-cinema is expected to rapidly video and audio (such as MPEG-2 and MP3) but also
expand as it has distinct advantages for the high quality include standards for indexing, filing and labelling material.
large images. It is already recommended for D-cinema and
Grass Valley have adopted it for HD compression in their
new Infinity range of cameras. www http://www.mpeg.org

As it does not analyse images block-by-block but in a

circular area-by-area pattern, there are no ‘blocky’
artefacts, instead problem areas tend to become a little
MPEG-2
softer – which is much less noticeable. JPEG 2000 ISO/IEC 13818-1. This is a video compression system
continues to improve as more bits are used for the images. primarily designed for use in the transmission of digital
As a result, at high bit rates of 200-300Mb/s HD and D- video and audio to viewers by use of very high compression
cinema images are displayed with ‘visually lossess’ quality. ratios. Its importance is huge as it is currently used for
It is also scalable, so image sizes different to the encoded nearly all DTV transmissions worldwide, SD and HD, as well
size, can be extracted directly without decoding. as for DVDs and many other applications where high video
compression ratios are needed.
The Profiles and Levels table (below) shows that it is not a
www http://www.jpeg.org
single standard but a whole family which uses similar tools
in different combinations for various applications. Although
all profile and level combinations use MPEG-2, moving
M-JPEG from one part of the table to another may be impossible
Motion JPEG refers to JPEG compression applied to without decoding to baseband video and recoding.
moving pictures. As the detail contained within each frame
varies, so some decision is required as to whether to use a Profile Simple Main 422P SNR* Spatial* High
4:2:0 4:2:0 4:2:2 4:2:0 4:2:0 4:2:0,4:2:2
constant bitrate scheme or constant quality. Level I, B I, B, P I, B, P I, B, P I, B, P I, B, P
High 1920x1152 1920x1152
See also: AVR, JPEG 80 Mb/s 100 Mb/s
High-1440 1440x1152 1440x1152 1440x1152
60 Mb/s 60 Mb/s 80 Mb/s

M-JPEG 2000 Main 720x570

15 Mb/s
720x576
15 Mb/s
720x608
50 Mb/s
720x576
15 Mb/s
720x576
20 Mb/s
JPEG 2000 used for moving pictures. Low 352x288 352x288
4 Mb/s 4 Mb/s

MPEG-2 profiles and levels

*SNR and Spatial are both scalable
Video Compression: Formats Understanding HD with Avid 21

Profiles outline the set of compression tools used. Levels

describe the picture format/quality from High Definition to
MPEG-4
VHS. There is a bit rate defined for each allocated profile MPEG-4 (ISO/IEC 14496) was developed by MPEG (Moving
/level combination. In all cases, the levels and bit rates Picture Experts Group) and is a wide standard covering
quoted are maximums so lower values may be used. many subjects but its importance in television production
Combinations applicable to modern HD are highlighted. is mostly related to its video compression scheme. MPEG-
4 Part 10, AVC (Advanced Video Coding) and H.264 all
MPEG-2 is deliberately highly asymmetrical in that decoding
refer to the same compression system. This is another
is far simpler than the encoding – so millions of viewers
DCT-based system that builds on MPEG-2 to produce a
enjoy reasonable prices while a few broadcasters incur the
more efficient codec again using intra and inter-frame
higher unit costs. Coding has two parts. The first uses DCT-
techniques. Coding is more complex than MPEG-2 but it
based intra-frame (I-frame) compression and application of
can produce extra data savings of around 30 percent – or
quantizing, to reduce the data – almost identically to JPEG.
more. Some of the latest television services are planned to
The second involves inter-frame compression – calculating
use MPEG-4. This is especially true with HD where more
the movement of macroblocks and then substituting just
bandwidth is required. It will enable the delivery of better
that information for the pictures between successive I-
image quality to viewers, or more channels to be delivered
frames – making a GOP. The movement is conveyed as
within a given bandwidth. It is said to be similar to, but not
motion vectors, showing direction and distance, which
the same as, WM 9.
amounts to far less data than is needed for I-frames.
Motion vector calculation is not an exact science so there
can be huge difference in quality between different MPEG
compressors. Decompression is deterministic so all
www http://www.chiariglione.org/mpeg

decompressors (decoders) should be the same.

The encoding process necessarily needs to look at several
VC-1
frames at once and so introduces a considerable delay. VC-1 is a video compression codec specification that is
Similarly, the decoder delays pictures. For transmissions this currently being standardised by SMPTE (SMPTE 421M) and
can add up to over a second. MPEG-2 is sometimes used on implemented by Microsoft as Windows Media Video
broadcast contribution circuits, this becomes noticeable (WMV) 9 Advanced Profile.
when news reporters appear to delay answering a question.
See: WMV 9
To fit HD video and audio down a transmission ‘data pipe’
requires very high compression. Uncompressed 10-bit HD
requires up to 1244Mb/s. But this is 10-bit data and WMV 9
sampled at 4:2:2. MPEG-2 is 8-bit sampled at 4:2:0 – Windows Media Video 9 is a video and audio compression
bringing the data down to 746Mb/s. However, the data system (codec) developed by Microsoft. It is said to be
pipes for ATSC (19.2Mb/s) or DVB (20Mb/s, depending on similar to MPEG-4 AVC and to have as good or slightly
channel width, parameters etc.) imply the need for around better performance giving lower data rates and claims to
40:1 compression. be a less complex process. It’s applications are seen as
See also: DCT, GOP, Intra-frame compression, Inter-frame compression. being for content delivery such as HD DVD.
Macroblock