1⁄4 1⁄4

1⁄4 1⁄4

Figure 12.4 An Interstitial chroma filter for JPEG/JFIF averages samples over a 2× 2 block. Shading represents the spatial extent of luma samples. The black dot indicates the effective subsampled chroma position, equidistant from the four luma samples. The outline represents the spatial extent of the result.

1⁄4 1⁄2 1⁄4

Figure 12.5 A cosited chroma filter for BT.601, 4:2:2 causes each filtered chroma sample to be positioned coincident – cosited – with an evennumbered luma sample.

1⁄8 1⁄4 1⁄8

1⁄8 1⁄4 1⁄8

Figure 12.6 A cosited chroma filter for MPEG-2, 4:2:0 produces a filtered result sample that is cosited horizontally, but sited interstitially in the vertical dimension.

Chroma subsampling filters

In chroma subsampling, the encoder discards selected colour difference samples after filtering. A decoder approximates the missing samples by interpolation. To perform 4:2:0 subsampling with minimum computation, some systems simply average CB over

a 2× 2 block and average CR over the same 2× 2 block, as sketched in Figure 12.4 in the margin. To interpolate the missing chroma samples prior to conversion back to R’G’B’, low-end systems simply replicate the subsampled CB and CR values throughout the 2× 2 quad. This technique is ubiquitous in JPEG/JFIF stillframes in computing, and is used in M-JPEG, H.261, and MPEG-1. This simple averaging process causes subsampled chroma to take an effective horizontal position halfway between two luma samples, what I call interstitial siting, not the cosited position standardized for studio video.

A simple way to perform 4:2:2 subsampling with horizontal cositing as required by BT.601 is to use weights of [1⁄4, 1⁄2, 1⁄4], as sketched in Figure 12.5.

4:2:2 subsampling has the advantage of no interaction with interlaced scanning.

A cosited horizontal filter can be combined with [1⁄2, 1⁄2] vertical averaging, as sketched in Figure 12.6, to implement 4:2:0 as used in MPEG-2.

Simple averaging filters like those of Figures 12.4, 12.5, and 12.6 have acceptable performance for stillframes, where any alias components that are generated remain stationary, or for desktop-quality video. However, in a moving image, an alias component introduced by poor filtering is liable to move at a rate different from the associated scene elements, and thereby produce a highly objectionable artifact. Highend digital video equipment uses sophisticated subsampling filters, where the subsampled CB and CR of a 2× 1 pair in 4:2:2 (or of a 2× 2 quad in 4:2:0) take contributions from several surrounding samples. The relationship of filter weights, frequency response, and filter performance will be detailed in Filtering and sampling, on page 191. These coefficients implement a high quality FIR filter suitable for 4:2:2 subsampling:

[-1, 3, -6, 12, -24, 80, 128, 80, -24, 12, -6, 3, -1]/256.

The video literature often calls these quantities chrominance. That term has a specific meaning in colour science, so in video I prefer the term modulated chroma.

See Introduction to composite NTSC and PAL, on page 135. Concerning SECAM, see SECAM, on page 126 of Composite NTSC and PAL: Legacy Video Systems.

Chroma in composite NTSC and PAL

I introduced the colour difference components PBPR and CBCR, often called chroma components. They accompany luma in a component video system. I also introduced UV and IQ components; these are intermediate quantities in the formation of modulated chroma.

Historically, insufficient channel capacity was available to transmit three colour components separately. The NTSC technique was devised to combine the three colour components into a single composite signal; the PAL technique is both a refinement of NTSC and an adaptation of NTSC to 576i scanning. (In SECAM, the three colour components are also combined into one signal. SECAM is a form of composite video, but the technique has little in common with NTSC and PAL, and it is of little commercial importance today.)

NTSC and PAL encoders traditionally started with R’G’B’ components. At the culmination of composite video, digital encoders started with Y’CBCR components. NTSC or PAL encoding involves these steps:

•Component signals are matrixed and conditioned to form colour difference signals U and V (or I and Q).

•U and V (or I and Q) are lowpass-filtered, then quadrature modulation imposes the two colour difference signals onto an unmodulated colour subcarrier, to produce a modulated chroma signal, C.

•Luma and chroma are summed. In studio video, summation exploits the frequency-interleaving principle. Composite NTSC and PAL signals were historically

analog. During the 1990s, digital composite (4fSC) systems were used; the 4fSC scheme is now obsolete. As I mentioned in Video system taxonomy, on page 94,

composite video has been supplanted by component video in consumers’ premises and in industrial applications. For further information, see Introduction to composite NTSC and PAL, on page 135.

The notation CCIR is often wrongly used to denote 576i25 scanning. The former CCIR (now ITU-R) standardized many scanning systems, not just 576i25.

Introduction to

In Raster scanning, on page 83, I introduced the concepts of raster scanning; in Introduction to luma and chroma, on page 121, I introduced the concepts of colour coding in video. This chapter combines the concepts of raster scanning and colour coding to form the basic technical parameters of the 480i and 576i SD systems. This chapter concerns modern systems that use component colour – digital Y’CBCR (BT.601), or analog Y’PBPR. In Introduction to composite NTSC and PAL, on page 135, I will describe NTSC and PAL composite video encoding.

Scanning standards

Two scanning standards are in use for conventional analog television broadcasting in different parts of the world. The 480i29.97 system is used primarily in North America and Japan, and today accounts for roughly 1⁄4 of all television receivers. The 576i25 system is used primarily in Europe, Asia, Australia, and Central America, and accounts for roughly 3⁄4 of all television receivers. 480i29.97 (or 525/59.94/2:1) is colloquially referred to as NTSC, and 576i25 (or 625/50/2:1) as PAL; however, the terms NTSC and PAL properly apply to colour encoding and not to scanning standards. It is obvious from the scanning nomenclature that the line counts and field rates differ between the two systems:

In 480i29.97 video, the field rate is exactly 60⁄1.001 Hz; in 576i25, the field rate is exactly 50 Hz.

Several different standards for 480i29.97 and 576i25 digital video are sketched in Figure 13.1 overleaf.

576i25 SCANNING

944

768

625 576

864

720

625 576

1135 4⁄625

948

625576 PAL

36 1⁄18

33:

See PAL-M, PAL-N on page 125, and

SECAM on page 126 of Composite NTSC and PAL: Legacy Video Systems. Consumer frustration with a diversity of functionally equivalent standards led to proliferation of multistandard TVs and VCRs in countries using these standards.

Analog broadcast of 480i usually uses NTSC colour coding with a colour subcarrier of about 3.58 MHz; analog broadcast of 576i usually uses PAL colour coding with a colour subcarrier of about 4.43 MHz. It is important to use a notation that distinguishes scanning from colour, because other combinations of scanning and colour coding are in use in large and important regions of the world. Brazil uses PAL-M, which has 480i scanning and PAL colour coding. Argentina uses PAL-N, which has 576i scanning and a 3.58 MHz colour subcarrier nearly identical to NTSC’s subcarrier. In France, Russia, and other countries, SECAM is used. Production equipment is no longer manufactured for any of these obscure standards: Production in these countries is done using 480i or 576i studio equipment, either in the component domain or in 480i NTSC or 576i PAL. These studio signals are then transcoded prior to broadcast: The colour encoding is altered – for example, from PAL to SECAM – without altering scanning.

ITU-R Rec. BT.601-5, Studio encoding parameters of digital television for standard 4:3 and widescreen 16:9 aspect ratios.

Figure 13.1 indicates STL and SAL for each standard. The SAL values are the result of some complicated issues to be discussed in Choice of SAL and SPW parameters on page 380. For details concerning my reference to 483 active lines (LA) in 480i systems, see Picture lines, on page 379.

Figure 13.2 above shows the standard 480i29.97 and 576i25 digital video sampling rates, and the colour coding usually associated with each of these standards. The 4:2:2, Y’CBCR system for SD is standardized in Recommendation BT.601 of the ITU Radiocommunication Sector (formerly CCIR). I call it BT.601.

With one exception, all of the sampling systems in Figure 13.2 have a whole number of samples per total line; these systems are line-locked. The exception is composite 4fSC PAL sampling, which has a noninteger number (11354⁄625) of samples per total line; this creates a huge nuisance for the system designer.

480i and 576i have gratuitous differences in many technical parameters, summarized in Table 13.1 overleaf.

†The EBU N10 component

analog interface for Y’PBPR, occasionally used for 480i, has

7:3 picture-to-sync ratio.

‡480i video in Japan, and the EBU N10 component analog interface, have zero setup. See page 381.

Table 13.1 Gratuitous differences. between 480i and 576i

2 EVEN

ODD 1

•

•

•

•

•

•

•

•

Different treatment of interlace between 480i and 576i imposes different structure onto the picture data. The differences cause headaches in systems such as MPEG that are designed to accommodate both 480i and 576i images. In Figures 13.3 and 13.4 below,

I show how field order, interlace nomenclature, and image structure are related. Figure 13.5 at the bottom of this page shows how MPEG-2 identifies each field as either top or bottom. In 480i video, the bottom field is the first field of the frame; in 576i, the top field is first. Figures 13.3, 13.4, and 13.5 depict just the image array (i.e., the active samples), without vertical blanking lines; MPEG makes no provision for halflines.

Figure 13.3 Interlacing in 480i. The first field (historically called odd, here denoted 1) starts with a full picture line, and ends with a left-hand halfline containing the bottom of the picture. The second field (here dashed, historically called even), transmitted about 1⁄60 s later, starts with a right-hand halfline containing the top of the picture; it ends with a full picture line.