Index Assignment for Progressive Transmission of Full Search Vector Quantization

Index Assignment for Progressive Transmission of Full Search Vector
Quantization

Abstract

The question of progressive image transmission for full search vector
quantization is addressed via codeword index assignment. Namely, we
develop three new methods of assigning indices to a vector
quantization codebook and formulate these assignments as labels of
nodes of a full search progressive transmission tree. This tree
defines from the bottom up the binary merging of codewords for
successively smaller-sized codebooks. The binary representation for
the path through the tree represents the progressive transmission
code. The methods of designing the tree which we apply are Kohonen's
self-organizing neutral net, a modification of the common splitting
technique for the generalized Lloyd algorithm, and, borrowing from
optimization theory, minimum cost perfect matching. Empirical
experiments were run on a medical image data base to compare the
signal-to-noise ratio (SNR) of the techniques on the intermediate as
well as the final images. While the neural net technique worked
reasonably well, the other two methods performed better and were close
in SNR to each other. We also compared our results to tree-structured
vector quantizers and confirmed that full search VQ has a slightly
higher SNR.