Weekly Projects, November 14

As in the last week, 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

Reproduce Sayood's example 8.5.2. Make expermiments aiming at establishing what happens to the rate (including overhead) and distortion (SNR) when we vary the parameters of the technique, i.e., 1) The block size, 2) the number of quantization levels. Does the distortion measured agree with the subjective distortion observed visually? What is the smallest number of bits per pixes we can use and still get a reasonable result? An extreme case of exampel 8.5.2 is the case of only two quantization levels. With only two levels it is feasible to send the parameters of the optimal quantizer (it is enough to send the two reconstruction levels). Implement this strategy and expermimentally evaluate the performance in terms of rate and distortion.

Try the same technique on sound files.

Project 2

Implement the Lloyd iteration finding the locally optimum quantization points for a given data set. Try it on actual images. Does the iteration converge? Fast? To the global optimum? Do the quantized images look well? Try a blocking strategy, as in example 8.5.2 but with (locally) optimal quantization points. What is the rate (including overhead) and distortion (SNR) achieved?

Try the same technique on sound files.

Project 3

Describe and implement a dynamic programming solution finding an optimum k-level scalar quantizer for a given data set. Try it on actual images. How fast is it? Try a blocking strategy, as in example 8.5.2 but with (locally) optimal quantization points. Do the quantized images look well? What is the rate (including overhead) and distortion (SNR) achieved?

Try the same technique on sound files.

Project 4

What is "dithering" and is it relevant for scalar quantization? Implement dithering and use it to lossily compress images. What is the achieved rate/distortion tradeoff? What is the subjective quality of the compressed images?