In a memoryless vector quantization system, each
image block is independently encoded as a corresponding
index and then an index table will be generated. In this
paper, we propose two novel schemes to compress the index
table without introducing extra encoding distortion. Both
our schemes exploit the characteristic that the blocks of
images are highly correlated to find the same or similar
index from the neighboring indices in the compression
process. To increase the compression, the principal
component analysis (PCA) technique is also employed to
sort the codewords in the codebook for minimizing the
difference of neighboring indices. In addition, our second
scheme excludes the repetitive indices from the search path
to further decrease the bit rate. Simulation results show
that our schemes are superior to SOC and traditional
memoryless VQ on the compression rate.
