Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 150

An ordered codebook is one where the blocks cropped from the “pink noise”
CA patterns are reordered according to an algorithm detailed in [96]. The ordering
process gives a significant reduction in the codebook-size thus improving the
compression rate at similar quality.
Figure 8.16 presents two examples of compressing the “Lenadetail” image
with different bitrates using CA-Q-E an d one example of compressing the “Cat”
image with a relatively low bitrate. In each case, the values of the w , T , and D
parameters are specified for each bitplane starting with the most significant at left.
The marking “-” indicates an ignored bit-plane.
Note the very high compression rate in Fig. 8.16a where a compression of more
than 160 times has been achieved (a rate of about 0.05 bits per pixel) with rea-
sonable blocking effects. The high frequency details in the original picture are still
present.
The picture in Fig. 8.16b has a quality comparable to the one obtainable using a
JPEG standard.
And finally, the “Cat” image compressed at 0.17 bpp has a reasonably visual
quality although the PSNR value is quite low. There is a simple explanation:
PSNR is lowered by the large presence of high frequencies (the grass). Although
the exact brightness of pixels in the grass region is altered, the overall high fre-
quency texture is well approximated so that the perceptual quality is not too much
affected.
A rate-distortion curve was determined for both CA-VQ and CA-VQ-E
schemes for both “Lena_detail” and “Cat” image. For reference, the same images
were encoded with the JPEG standard using the JPEG converter (imwrite func-
tion) from the Matlab environment with different quality factors. The resulting
distortion curves are plotted in Fig. 8.17. In JPEG encoding the quality degrades
suddenly for low bitrates, while the proposed CA-VQ performs better. The quality
expressed as PSNR is lower than for PEG when bitrates are higher than 0.25 bpp
but the difference in terms of perceptual quality are negligible.
The above statements hold for both “Lena_detail” and “Cat” images and can be
generalized for arbitrary images as well. The PSNR is lower for the “Cat” image
mainly because of the high frequency content due to the grass in that scene. A
small PSNR is also obtained when JPEG encoding standard is used.
Therefore we conclude that our compression scheme is particularly well suited
for low bitrates (i.e. high compression) although its implementation simplicity
(compared with JPEG) may impose it even for higer bitrates, in applications
requiring economic implementation.