Lossless Image Compression - Introduction to Data Compression - page 190

Assuming that this approximation was sufficient to let the user decide whether a particular
image was the desired image, the time required now to look through 30 images becomes on
the order of seconds instead of the 15 minutes mentioned earlier. If the approximation using a
block size of 8

×

8 does not provide enough resolution to make a decision, the user can ask for
a refinement. The transmitter can then divide the 8

× ×

4 blocks. The pixel
at the upper-left corner of the upper-left block was already transmitted as the representative
pixel for the 8

8 block into four 4

4 blocks.
This takes a little more than a second, so even if the user had to request a finer approximation
every third image, this would only increase the total search time by a very small amount. To
see what these approximations look like, we have taken the Sena image and encoded it using
different block sizes. The results are shown in Figure 7.3 . The lowest-resolution image, shown
in the top left, is a 32

× 8 block, so we need to send three more pixels for the other three 4 ×

× 32 image. The top-right image is a 64 × 64 image. The bottom-left

image is a 128 × 128 image, and the bottom-right image is the 256 × 256 original.

F I GU R E 7 . 3 Sena image coded using different block sizes for progressive trans-

mission. Top row: block size 8 × 8andblocksize4 × 4. Bottom row:

block size 2 × 2 and original image.