visibility. However in general, DCT-based processes cannot be used for true lossless coding.
than the DCT-based processes for distortions near and above the threshold of visibility.
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.
techniques. Among these are
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.
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.
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.
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.
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.
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.
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.
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.