Bin
Yu
3. LOSSY DATA
COMPRESSION

3.2 Adaptive Quantization
My interactions with
Martin Vetterli's
group in the EE department at Berkeley
started in 1994.
The Ph.D. thesis of Antonio Ortega
with Vetterli in 1994 contained a novel algorithm which
modifies the quantization levels of the incoming real data
based on quantized past data. The need for this type
of algorithm can be envisioned in the transmission of a large quantity
of images over the internet where both ends have free
access only to quantized past data.
B. Yu,''
A Statistical analysis of adaptive scalar quantization
based on
quantized past data,''
Submitted
IEEE Trans. on Information Theory, 1996.
analyzes the Ortega-Vetterli (OV) adaptive algorithm
in Ortega's thesis.
In particular, I characterize
the approximate density to which the algorithm converges
and I explain why the algorithm works and under what conditions it can
be expected to work.
Based on Bennett's
distortion integral, it is argued that
estimating the interval partition
of the optimal quantizer
is an easier problem than density estimation and that
the OV algorithm performs well
for this interval partition estimation.
Procedures are also proposed to improve the OV reproduction
level estimates in the low resolution case.
Recent subband image coding techniques owe much of their
success to an effective use of adaptive
quantization and adaptive entropy coding. It is often the case that
adaptive quantization is achieved by defining a discrete
set of quantizers from which one is chosen for a given set
of coefficients. This type of forward adaptation thus requires
that overhead information (the choice of quantizer) be sent
to the decoder. Then, the quantized coefficients are transmitted using
adaptive entropy coding, typically through backward adaptive
arithmetic coding.
Y. Yoo, A. Ortega, and B. Yu,''
Adaptive quantization of image subbands with efficient
overhead rate selection,''
In
Proceedings of IEEE International
Conference
on Image Processing, Lausanne, Switzerland, 1996.
shows that
a combination of forward and backward adaptation methods can be
used to update the quantizers thus reducing the overhead
requirements while still providing good performance.
Specifically, we present
an algorithm where each coefficient is classified into several classes
based on the past quantized
data and where the quantizer to be used for each class
can itself be adapted on the fly.