2. Digital Video

Digital Video (slide 7)

By moving on from sound to image, we also move from an one-dimensional to a two-dimensional physical reality, from a line to a surface.

A digital image is usually represented by a rectangular matrix of values, a raster. A single element in this image matrix is called picture element or in short pixel. In a colour system, a pixel includes the information for all colour components.

I should clarify that in today’s digital world we have only square pixels. In the past, during the transition from analogue to digital, there were also rectangular pixels. Yet today, if you digitise an analogue film or video, you should always choose square pixels.

The first point in the summary, the resolution, is about the quantity of pixels, about the horizontal and vertical size of the matrix; and all the following mentioned points are about the quality of the pixels in the matrix.

Resolution (slide 8)

The resolution measures the quantity of the image units, of the pixels. It tells how many pixels the image measures horizontally and vertically. Often people thinks that this value determines a digitisation, but this is not true: resolution is an important factor, but resolution says not at all everything about a digital file. The image resolution is similar to the sampling rate for the sound.

Some current resolutions are:

  • SD is the old television format. For NTSC the matrix measures 640 x 480 square pixel1 (Standard Definition 480 interlaced) and for PAL and SECAM the matrix measures 768 x 576 square pixel (Standard Definition 576 interlaced).
  • For historical reasons, the modern television comes in two sizes: HD 720p with 1280 x 720 pixel (High Definition 720 progressive) and HD 1080i with 1920 x 1080 pixel (High Definition 1080 interlaced). Those formats are mainly used for diffusion, and sometimes for production, but cannot be recommended for digitisation and for archiving.
  • 2K with 2048 pixel horizontally and HD 1080p with 1920 x 1080 pixel (High Definition 1080 progressive) can be used for digitisation of analogue film.2 Today HD is often used in short to refer to HD 1080p.
  • 4K with 4096 pixel horizontally and UHD-1 (Ultra High Definition) with 3840 x 2160 pixel. This is the double horizontally and the double vertically compared to 2K and HD, that means four times the resolution – and four times the needed storage!
  • And doubling again: 8K with 8192 pixel horizontally and UHD-2 with 7680 x 4320 pixel.3
Bit Depth (slide 9)

Bit depth for image is very similar to quantisation for sound. Usually those values are given for one colour channel, and usually there are three colour channels.

Typical bit depth values and uses are:

  • 8 bit for computer display, digital photography and SD television;
  • 10 bit for professional HD video, HD television and cinema;
  • 12 bit for the majority of today’s sensors;
  • 16 bit for special effects (of course this is very useful also for the film restoration).
Linear vs. Logarithmic (slides 10–11)

The colour tones can be represented according to a linear scale or to a logarithmic scale. The linear scale represents the physical reality of the wavelength; the logarithmic scale represents the human perception of colours. The so called medium grey is 18% on a linear scale and 50% on a logarithmic scale. The conversion between linear scale and logarithmic scale is done by a transfer function called with one single parameter γ (gamma).

Typical γ values are:

  • γ = 2.6 for a film projection in a cinema theatre;
  • γ = 2.4 for working in a professional HD postproduction studio with controlled light;
  • γ = 2.2 for watching television in an usual room at home (according to Rec. 601).
Colour Model (slide 12)

The colour space XYZ was defined by the Commission internationale de l’Éclairage (CIE, International Commission on Illumination) in 1931 and reviewed in 1964. It gives a precise mathematical description of the physical reality. All the colour models are connected to this colour space.

There are two main colour models used in cinema: the additive and the subtractive. An additive colour model is based on transmitted light and can be linear (RGB) or logarithmic (R′G′B′). A subtractive colour model is based on reflected light and can also be linear (CMY) or logarithmic (C′M′Y′).4 RGB is also used for computer screens and CMYK for inkjet printers, where K designs the additional black cartridge.

Three different colour models were used for the analogue composite television: Y′IQ for NTSC, Y′UV for PAL and Y′DBDR for SECAM.

In today’s digital video the colour model Y′CBCR is almost always used.5 Therefore there are also other colour models, for example Y′COCG which is internally used by some codecs, such as H.264.

The colour model Y′PBPR has been very important in the past, during the transition from analogue component television to digital component television. It permits to send over an analogue network content that has been created digitally according to Y′CBCR.

Example: Beamer DLA-HD1 from JVC (slide 13)

This model comes with a large connectivity as shown on the rear.

Example: RGB24 (slide 14)

RGB24 is a 24 bit colour schema. Every pixel is represented by 24 bit (or 3 byte), while the first 8 bit represent the red value, the 8 bit in the middle represent the green value and the last 8 bit represent the blue value. By using those 24 bits, 2 power 24 equals 16 777 216 different colours can be digitally encoded. This is the “16 million color” setting in the computer screen parameters.

Chroma Subsampling (slides 15–19)

Chroma subsampling is usually not called compression, but scientifically it is one, and even a lossy one.

  • 4:4:4 is the full information;
  • 4:2:2 reduces the volume and the information by 1/3;
  • 4:2:0 and 4:1:1 reduce the volume and the information by 1/2.
Illuminant (slides 20–21)

The CIE standard illuminant is the point in the XYZ colour space where the so called white point is located. For example is gives the exact coordinates of the RGB24 value (11111111, 11111111, 11111111).

D65 denotes the bright light at noon. Some consider that this white is in fact a little too bluish.

The values shown as an example are for HDTV. Not the full human visible colours can be represented, but only the colours inside the triangle determined by the primaries red, green and blue.


This is also the size of VGA (Video Graphics Array).
When digitising film in 2K, then a resolution of 2048 pixel horizontally is considered. Yet when projecting the same film in a theatre as a 2K DCP “flat” or for screening on HD television, then 1080 pixel vertically are considered. Therefore, if an old image with aspect ratio 4:3 is shown on a modern screen with an aspect ratio of 16:9, then only 1440 pixels can be used horizontally, and the right and left side of the screen should stay black (pillarbox). Therefore the resolution is approximately 1.4K in reality.
UHD-2 is supposed to be used for the 2020 Olympic Games in Tokyo. That should be the first main event produced and screened live in this format.
Scientifically speaking, this is a multiplicative colour model, but that naming is not used at all in the literature.
Often computer people call this colour model YUV, which is actually composite analog PAL, and video people use often YCbCr and sometimes Y′CbCr.