Video compression

The aim in video compression is to remove those parts of the signal which we do not essentially need.

The reason why video compression works so well is that video data is described as being ‘very redundant’, due to the way the human psychovisual system operates. There are three types of redundancy in video:

Spectral redundancy

Bright pixels tend to be bright in all three colours, red, green and blue, and therefore there is said to be spectral redundancy – the similarity between colour values at any one point in the picture.

Earlier we mentioned the ability to fool the human brain into thinking that it is seeing a true representation of the original picture. This extends to how the brain is able to distinguish the brightness (luminance) and colour (chrominance) detail of a picture.

The human eye is much better at distinguishing brightness differences than colour differences. This can be used to an advantage in conveying colour information, as there is less precision required – a higher level of precision would be ‘wasted’ as the brain is unable to make use of the additional information and distinguish the difference.

When digitizing a video signal, fewer samples are required to convey the colour information, and fewer samples means less bandwidth. Typically, for every two luminance samples there is only one chrominance sample required.

Spatial redundancy

In the digital domain the removal of further redundant information is carried out in two dimensions within each frame of picture, described as spatial redundancy (i.e. relating to a frame of space).

Spatial redundancy is the relationship each pixel has with the neighbouring pixels around it – more often than not, two neighbouring pixels will have very nearly the same brightness and colour values.

Consider one frame of a picture signal. Suppose that within this frame there is a notable area of the picture that has either the same colour or brightness content – or both.

The luminance and/or chrominance would be the same across this similar area, and therefore instead of sending the same numbers for each and every sample, one number could be sent representing a block of sample points in an area where the information content remains the same.

Hence we have found some spatial redundancy, and this reduces the amount of data that has to be sent.

Temporal redundancy

In a TV picture, there is a sense of motion because there is a difference in the position of objects between one frame and the next.

Objects that have moved in the time between one frame and the next often may not significantly change shape, so by sending the same information from one frame to the next, some redundancy in the data from one frame to the next can be removed. This redundancy between successive frames is described as temporal redundancy (temporal meaning it relates to time).

Consider two frames of a picture signal. In the first frame, there is an object that is also present in the second frame – for example, let us presume that the first frame has a building in it.

Unless there is a shot change or a rapid camera pan, the building will be the same shape and present in the next frame (though not necessarily in the same position if the camera has moved slightly).

The block of data samples that described the building in the first frame could therefore be repeated in the second frame. In fact, most pictures shot in a single sequence are remarkably similar from one frame to the next, even in a moderately quickly moving sequence – only 1/25 (PAL) or 1/30 (NTSC) second has elapsed, which is much faster than the blink of an eye!

Therefore there is a significant scope for data reduction by simply reusing sets of data sent for the previous frame, albeit that the blocks of data may have to be mapped (moved) to a different position in the frame.