MPEG-2 Video Coding - Digital Video and Audio Broadcasting Technology - page 123

zation must be changed only a little or not at all and in the case of fine im-
age structures, the quantization is reduced more, meaning that the quanti-
zation factors increase in the direction of finer image structures. Due to the
quantization, many values which have become zero are obtained as the
fineness of the image structure increases, that is to say in the direction of
higher frequency coefficients.

f(x,y)

Fig. 7.17. 8 x 8 pixel block

55 70 92 111 116 108 94 83

70 81 94 103 101 87 70 58

16 81 42 56 62 62 57 52

34 36 38 39 37 33 29 26

67 63 57 52 50 50 51 53

46 40 33 28 32 42 55 64

46 32 10 -9 -16 -13 -3 5

24 10 -9 -23 -24 -13 5 18

Fig. 7.18. Subtracting 128

These values can then be coded in a special space-saving way. How-
ever, the characteristic recovered by decoding at the receiving end after the
quantization then no longer corresponds perfectly to the original curve
(Fig.7.15.) and exhibits quantization errors.
In practice, however, the coding in JPEG and MPEG is two-dimensional
transform coding. For this, the picture is divided into 8 x 8 pixel blocks
(Fig.7.8.). Each 8 x 8 pixel block (Fig.7.16.) is then transformed into the
frequency domain by means of the two-dimensional Discrete Cosine
Transform. Before that is done, the value 128 is first subtracted from all
pixel values in order to obtain signed values (Fig.7.17.).