Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 149

are convenient to be used in compression schemes with different block sizes per
bitplane or they can be optimally tuned to certain classes of images. Among the
above list, a finer investigation led to the following ( ID,T,PSNR ) pairs as good
choices among the list: (219, 100, 23.6 dB), (131, 50, 23.8 dB), (147, 20,
23.6 dB), (852, 200, 23.4 dB). Fine tuning of T may slightly improve the PSNR.
The codebook patterns obtained for ID = 131 after different numbers of itera-
tions are depicted in Fig. 8.15.
Fig. 8.15. Codebooks for ID = 131 and their power spectra for different numbers of itera-
tions
8.4.3 Results and Comparison with Other Compression Standards
In order to compare the CA-VQ scheme with other compression standard, a large
variety of images was considerded. Here we will detail the results for two of them:
The first, entitled “Lena_detail”, is an image giving a good reconstruction error
(large PSNR), and is often used in image processing algorithms. In order to ob-
serve the details we cropped a block of 192×192 pixels containing the face from
the original 512 × 512 pixels “Lena” image. The second image, “Cat_in_the_grass”
or simply “Cat” comes from a photo of the author (here in 624 × 480 pixel resolu-
tion) and has a rich content of high frequencies giving the lowest PSNR among the
images used in tests.
Variable block-sizes were used with an independent optimization of the code-
books per each bitplane, to achieve the best performance for a given compression
rate. By changing the block-size profile one can easily achieve a desired bitrate. In
all cases the CA used as a codebook generator has ID = 131 while the codebook
optimization is done only with respect to the T (number of CA iterations) and D
(number of code-words in the ordered codebook) for a given block-size w.