Multiple Description Vector Quantization - Animated Partitions
This page provides links to animations to supplement Figure 6 of
-
V.K. Goyal, J.A. Kelner, and J. Kovacevic,
"Multiple Description Vector Quantization with a Coarse Lattice,"
to appear in IEEE Trans. on Information Theory, March 2002.
(abstract)
(PDF)
These animations show partitions of two-dimensional real space obtained
with the modifications of MDLVQ first proposed in
-
J.A. Kelner, V.K. Goyal, and J. Kovacevic,
"Multiple Description Lattice Vector Quantization:
Variations and Extensions,"
Proc. IEEE Data Compression Conf., pp. 480-489, March 2000.
(abstract)
(PDF)
(postscript)
and developed more fully in the first paper listed above.
The starting lattice is the two-dimensional hexagonal lattice
and sublattice indices of 7 and 13 are used.
The partitions change as the parameter p is varied from 0 to 1.
When this parameter is interpreted as the probability of losing
one description, the partitions are optimized to minimize the
expected squared-error distortion.
About the files
These animations were produced with MATLAB. They are provided
as gzipped MATLAB movies and as .avi files.
Many media players will play .avi files.
To use a .mat.gz file, uncompress the file, load it into MATLAB,
and use the "movie" command.
For example, after uncompressing the first file, use the following
commands in MATLAB:
load movieLattice7
figure; axis off; axis equal; movie(M,0,5)
The animations (in the same order as the columns of Figure 6)
- Index 7 with modified encoding
- Index 7 with modified encoding and centroid decoding
- Index 7 with optimized placement of the central decoder codebook
- Index 13 with modified encoding
- Index 13 with modified encoding and centroid decoding
- Index 13 with optimized placement of the central decoder codebook
WARNING: The .mat files will expand by about 25 times when uncompressed.
Vivek Goyal / v.goyal@ieee.org
Last updated 15 Aug 01
Copyright © Vivek Goyal 2001. All rights reserved.