Databases Reference
In-Depth Information
(a)
Use the two quantizers to quantize a sequence of 10,000 zero mean unit variance
Laplacian random numbers. Using either the mean squared error or the SNR as a
measure of performance, compare the performance of the two quantizers.
(b)
Use the two quantizers to quantize the Sinan image. Compare the two quantizers us-
ing either the mean squared error or the SNR and the reconstructed image. Compare
the difference between the performance of the two quantizers with the difference
when the input was random.
7.
Design one-, two-, three-, four-, and eight-dimensional quantizers at the rate of 1 bit/sample
using the LGB algorithm (you can use
trvqsplit
), for the uniform, Gaussian, and Laplacian
distributions. Study how your design algorithm performs and suggest ways of improving
it.
Use the randomnumber generator programs to design and test the quantizers (
rangen.c
).
8.
The purpose of this project is to have you familiarize yourself with vector quantization,
in particular with vector quantization of images. The programs you will use include
trvqsp_img
,
vqimg_enc
, and
vqimg_dec
.The
trvqsp_img
program designs
vector quantization codebooks for images using the splitting approach. The programs
vqimg_enc
and
vqimg_dec
are the vector quantization encoder and decoder for
images. The indices generated by the encoder are coded using fixed-length codes.
Experiment with different ways of selecting vectors and examine the effect of the training
set used for obtaining a codebook on the VQ performance. Suggest some ways of
resolving the mismatch problem between the image being encoded and the codebook.
