1 ‘hip 7» Ont compression
Space dana: ese dona
Tage day Eel
fre ta
Figure 7-11 Relationship between the two-dimensional space and frequency
domains
‘The coefiicient S., corresponds to the portion where the frequency is zero in both
«dimensions. 1is also known as the DC-coefficient (OC voltage portion) and determines
the fundamental color of all 64 pixels of the data unit, The other coefficients are called
AC-coefficients (analogous to the AC voltage pontion). For instance, $;, represents the
highest frequency that eceurs in the horizontal direction, that is, the closest possible
separation of vertical lines in the 88 data unit. Sy epresents the highest frequency in
the vertical dimension, that is, the closest possible separation of horizontal lines. Sy
indicates the highest frequency appearing equally in both dimensions, Its absolute value
is greates ifthe original data unit contains as many squares as possible, thats, iit eon-
s solely of 1x1 squares, Accordingly, for example, $35 will be maximal if a block
consists of 16 squares of 434 pixels, Taking a closer look atthe above formula, we ree-
‘ognize that the cosine expressions depend only upon x and u, or upon y and y, but not on
Six Therefore, these expressions represent constants that do not need to be recalculated
over and over again, There are many effective DCT techniques and implementations
[DG90, Fei90, Hou88, Lee84, LF91, SH86, VNS4, Vet85]
Por later reconstruction of the image, the decoder uses the Inverse DCT (IDCT),
‘The coefficients S,, must be used for the calculation:
sy = BS Severn OE Dihagg IED
wher by 0; otherwise ¢,
Ifthe DCT, as well asthe IDCT, could be calculated with fall precision, it would
be possible to reproduce the original 64 pixels exactly. From a theoretical point of view,
DCT would be lossless in this ease. In practice, precision is limited and DCT is thus
lossy. The JPEG standard does not define any specific precision. It is thus possible that