ISO/IEC 10918-1 : 1993(E)
Domain of applicability of DCT and spatial coding techniques
The DCT coder is intended for lossy coding in a range from quite visible loss to distortion well below the threshold for
visibility. However in general, DCT-based processes cannot be used for true lossless coding.
The lossless coder is intended for completely lossless coding. The lossless coding process is significantly less effective
than the DCT-based processes for distortions near and above the threshold of visibility.
The point transform of the input to the lossless coder permits a very restricted form of lossy coding with the "lossless"
coder. (The coder is still lossless after the input point transform.) Since the DCT is intended for lossy coding, there may
be some confusion about when this alternative lossy technique should be used.
Lossless coding with a point transformed input is intended for applications which cannot be addressed by DCT coding
techniques. Among these are
true lossless coding to a specified precision;
lossy coding with precisely defined error bounds;
hierarchical progression to a truly lossless final stage.
If lossless coding with a point transformed input is used in applications which can be met effectively by DCT coding, the
results will be significantly less satisfactory. For example, distortion in the form of visible contours usually appears when
precision of the luminance component is reduced to about six bits. For normal image data, this occurs at bit rates well
above those for which the DCT gives outputs which are visually indistinguishable from the source.
Domain of applicability of the progressive coding modes of operation
Two very different progressive coding modes of operation have been defined, progressive coding of the DCT coefficients
and hierarchical progression. Progressive coding of the DCT coefficients has two complementary procedures, spectral
selection and successive approximation. Because of this diversity of choices, there may be some confusion as to which
method of progression to use for a given application.
Progressive coding of the DCT
In progressive coding of the DCT coefficients two complementary procedures are defined for decomposing the 8
coefficient array, spectral selection and successive approximation. Spectral selection partitions zig-zag array of DCT
coefficients into "bands", one band being coded in each scan. Successive approximation codes the coefficients with
reduced precision in the first scan; in each subsequent scan the precision is increased by one bit.
A single forward DCT is calculated for these procedures. When all coefficients are coded to full precision, the DCT is the
same as in the sequential mode. Therefore, like the sequential DCT coding, progressive coding of DCT coefficients is
intended for applications which need very good compression for a given level of visual distortion.
The simplest progressive coding technique is spectral selection; indeed, because of this simplicity, some applications may
choose despite the limited progression that can be achieved to use only spectral selection. Note, however, that the
absence of high frequency bands typically leads for a given bit rate to a significantly lower image quality in the
intermediate stages than can be achieved with the more general progressions. The net coding efficiency at the completion
of the final stage is typically comparable to or slightly less than that achieved with the sequential DCT.
A much more flexible progressive system is attained at some increase in complexity when successive approximation is
added to the spectral selection progression. For a given bit rate, this system typically provides significantly better image
quality than spectral selection alone. The net coding efficiency at the completion of the final stage is typically comparable
to or slightly better than that achieved with the sequential DCT.
Hierarchical progression permits a sequence of outputs of increasing spatial resolution, and also allows refinement of
image quality at a given spatial resolution. Both DCT and spatial versions of the hierarchical progression are allowed, and
progressive coding of DCT coefficients may be used in a frame of the DCT hierarchical progression.
The DCT hierarchical progression is intended for applications which need very good compression for a given level of
visual distortion; the spatial hierarchical progression is intended for applications which need a simple progression with a
truly lossless final stage. Figure K.6 illustrates examples of these two basic hierarchical progressions.
CCITT Rec. T.81 (1992 E)