3.6.2 DCT. Discrete Cosine Transformation
The Discrete Cosine Transformation is a variant of the Fourier-Transform
(Fourier-Series). The basic idea behind these transformations is that
any (periodic) function can be decomposed in sine and cosine functions
with frequencies being multiples of the given periodicity.
Consider the symmetric -matrix
Its Eigenvectors are
for
with Eigenvalues
for ,
since
for
we have
using
Since is symmetric, and the sequence of Eigenvalues is strictly monotone increasing,
we conclude that the Eigenvectors are orthogonal, thus
any vector
can be reconstructed from the projections
via
The norms are
and
for , since
using
and
.
So
defines an ISOMETRY
, i.e. is length-preserving
with respect to the euclidean norm, since
The 2-dimensional DCT is now defined by applying the 1-dimensional DCT
to rows and columns, i.e.
Thus with respect to the -norm we get for the 2-dimensional DCT:
Note that here we used a shift for the original data
to
.
See also
www.ztt.fh-worms.de/.../node34.html
Andreas Kriegl 2003-07-23