Fwlbg_adap Construct a complete set of codebook sequences adapted to vector quantization of image wavelet transform
Fwlbg_adap [-b1 BiFilt1] [-b2 BiFilt2] [-o OrthoFilt] [-e EdgeFilt] codebook trainimage1 [trainimage2 [trainimage3 [trainimage4]]]
This macro generates a complete set of codebook sequences for the quantization of image wavelet transforms. It is based on the fwlbg_adap module, but is simpler to use.
One only has to specify the prefix codebook of all generated codebook sets and at least one (maximum four) input image for the training sets. The macro generates codebook sequences CodeBook1, CodeBook2, CodeBook3, ResCodeBook1, ResCodeBook2, ResCodeBook3, ResResCodeBook1 and ResResCodeBook2, which can be used directly as input to the compression module fwvq. These are put in the files codebook.cb, codebook_x.cb, codebook_y.cb, codebook_q.cb, codebook_xq.cb, codebook_yq.cb, codebook_qr.cb and codebook_xqr.cb respectively. However, if one uses the Fwvq and Fwivq macros, one does not have to bother about these files. It suffices to put the same prefix name codebook as argument of Fwvq and Fwivq (see documentation).
The default wavelet transform is a biorthogonal one using 7/9 Cohen-Daubechies filter pair (see [CDF92]). The filters are read in the h/sd07.ri and htilde/sd09.ri which must be available either in the MegaWave2 data directory or data/quantization subdirectory. However one can use other filter pairs (using -b1 and -b2 options) or switch to orthogonal transform, using -o option (possibly with the -e option if special edge processing is requested, see owave2 module documentation for details). The normalization mode is the default one.
Codebook sequences are generated for details from level 1 to 4, and for average at level 3 and 4. The size of blocks is 4×4 at levels 1 and 2, 2×2 at levels 3 and 4.
The overlapping option is activated, as well as the default decimation stop (at level 2) option.
The maximum size of codebooks are
The threshold value options are set to 3.0 and 1.0.
This option setting is well adapted to the construction of a codebook set
from 2,3 or 4
512×512 images. It enables to get rather high bit rates
when using fwvq module or Fwvq macro with the generated codebook
set. However, the running time might be very long (several days!).
In order to perform a quick test, one has to modify the maximum sizes
of codebooks directly in the macro file. Typically, one should put
sizes not greater than 16. However, one won't be able to get
very high bit rate compression when using these codebooks.
Last Modification date : Feb 11 2002