In this problem we will compare the performance of a 16-dimensional pyramid vector quantizer and a 16-dimensional LBG vector quantizer for two different sources. In each case the codebook for the pyramid vector quantizer consists of 272 elements:
• 32 vectors with 1 element equal to ibA, and the other 15 equal to zero, and
• 240 vectors with 2 elements equal to zLA and the other 14 equal to zero. The value of A should be adjusted to give the best performance.
The codebook for the LBG vector quantizer will be obtained by using the program trvqsp_img on the source output. You will have to modify trvqsp_img slightly to give you a codebook that is not a power of two.
(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 using 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.