readme2

来自「Vector Quantization压缩算法」· 代码 · 共 42 行

EXAMPLE:

Systematic joint best-basis selection and bit allocation for
one dimensional data at 2 bits per sample.

Given five training images are 512X512 in dimension, output, as
doubles, the wavelet coefficients from a one-dimensional two level
wavelet packet transform into files (subband, subband.1, subband.2,
subband.1.1, subband.1.2, subband.2.1, and subband.2.2) where subband
is the original data, subband.1 (subband.2) is the lowpass (highpass)
subband, subband.1.1 is the lowpass subband of the lowpass subband,
and so on.

Use block and tsvq, both compiled to handle double DATATYPE, to make
TSVQ codebooks for each training data file, and then make PTSVQ's.  A
shell script is recommended to make this much easier to handle.

e.g.	block -i subband.1.2 -o train.1.2 -r 640 -l 128 -h 2 -w 2
	tsvq -t train.1.2 -c codebook.1.2 -s statfile.1.2 -d 4 -r 8 > tsvq.out
	prune -c codebook.1.2 -s statfile.1.2 -o nested.1.2 >& prune.out.1.2

Note that the image subband dimension for the second level of
decomposition is 128 by 128, but as there were 5 training images,
there are 5*128 = 640 rows.

Once the above step is done there should be files (prune.out,
prune.out.1, prune.out.2, prune.out.1.1, prune.out.1.2, prune.out.2.1,
and prune.out.2.2).  These contain all of the rate-distortion
characteristics of all the PTSVQ's.  Systematic joint best-basis
selection and bit allocation is done as:

wpt_bitalloc -c prune.out -l 2 -d 2 -r 2 > wpt_bitalloc.out

The output of wpt_bitalloc consist of several scripts to assist
encoding and reconstruction of actual test data.  These are examples,
and should be modified as needed.  For more details, see the
documentation in the code and the tsvq code.

Questions and comments should be sent to
Jill R. Goldschneider
[e-mail]