To test our three algorithms we try to compress the standard
images *Lenna *and *Clown. *(See Figures
and ).
For the DCT based compression algorithm we determined values for
the PSNR of the compressed images for various compression rates in order
to measure the performance of our coders. In
order to do this, we had to generate a bit allocation mask for each
desired compression rate. For the case of the wavelet transformation
we developed a function (mask.m) to create masks at the compression
rates of: 1, 75, .675, .5, .25, .125, and .0625. We used a slightly
more ad hoc method for developing masks in the case of the DCT. Also,
for the Haar DWT, we stopped transforming the input image at a minimum
block size of
.
We
then measured the MSE of the compressed images, and calculated the
PSNR from that. We based The bits-per-pixel for each image on the
file size of the gzipped compressed image. The results of this
experiment performed on both the lenna image and the clown image can
be seen in Figure .

For EZW algorithm the quantization is implicitly implemented after each subordinate pass, so the quantization table is actually related to the size of the bit stream for reconstruction. To do the experiment we first generate a bit stream which can be used for perfect reconstruction of the image. Then based on our desired compression ratio we can determine how many bits we need for reconstruction. Then we truncate it to the right size, and the image is reconstructed based on the truncated bits. Since the EZW is a progressive coding algorithm, the bit stream of any size will construct an approximation of the original image.