The quantisation process refers to the DCT coefficients and is performed in order to both remove the subjective redundancy and control the compression factor.
The setting of the quantisation parameters is a key point for the quality of the coder.

To remove the subjective redundancy some quantisation matrices are used.
A couple of default matrices are given, for Intra and Non-Intra blocks, but other matrices can be transmitted at the sequence layer.
Each element of a quantisation matrix refers to a coefficient of the Discrete Cosine Transform and it should be proportional to the subjective redundancy of that coefficient.
In general lower spatial frequencies coefficients are better quantised than higher frequencies ones.
With 4:2:2 or 4:4:4 data four matrices can be used, distinguishing between Luminance and Chrominance.

In order to perform the bit-rate control or, more in general, to achieve an high enough compression factor , it is often necessary to quantise the DCT coefficients more heavily.
So a quantiser_scale_code is transmitted at each slice and may be transmitted also at each macroblock.
The quantiser_scale_code is a pointer to the value of the quantiser_scale, which is applied to any DCT coefficient, except the DC coefficient of Intra blocks.
Two different tables can be chosen at the picture layer setting the q_scale_type flag.

The Intra DC coefficients are quantised in a different manner to all other coefficients.
The number of bits of precision (from 8 to 11) is set by the intra_dc_precision field at the picture_layer