Weekly Projects, November 21
As in the last weeks, choose just one of the projects. You
are also most welcome to come up with alternative project ideas
yourself, as long as they are related to the issues we are
currently
discussing.
Project 1
Many tools (like xv) implements lossy jpeg with a possibility of
controlling the quality/file size tradeoff. Run such a program with
many settings of the trade-off for a fixed image, such as Lena.
Plot the SNR as a function of the rate in bits/pixel. Can you verify
the 6 db rule?
Project 2
Use uniform scalar quantization combined with entropy coding to encode
images and sound. Plot the SNR as a function of the rate in
bits/sample or bits/pixel. Can you verify the 6 dB rule?
Project 3
Implement differential coding to do lossy compression using
model coefficinet minimizing sum-of-prediction-errors-squared.
Combine with a quantizer and use it to compress benchmark sound (or image,
if you use a two-dimensional model) files. Plot the SNR as a function
of the rate (and comment on preceived quality). Does the performance
improve if you split up the file
into several smaller blocks and compute an optimal model for each
block? Include the overhead of describing the model(s) in the estimate
of the code word length. Note that the model parameters must also
be quantized somehow!
Project 4
Present adaptive differentital coding
(Sayood, pages 273-275). Perform
experiments with adaptive differential coding similar to the
ones of Project 3.
Project 5
Implement versions of Delta modulation. Plot the SNR as a
function as a function of the rate. Comment on the perceived
quality. Does it make a difference whether
you apply a low pass filter or not to the reconstructed signal?
Does your choice of adaptation scheme make a difference?