pSQ - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 63

FIGURE 3 Examples of recovered images of F-16 compressed at 0.4 bpp. (a) JPEG2000
PSNR = 25.12 dB. (b) JPEG2000-F-pSQ PSNR = 24.57 dB.
Figure 4 shows examples of recovered images of Baboon compressed at 0.59, and 0.45 bits per
pixel by means of JPEG2000 (a) without and (b) with F-pSQ. In Figure 4(a) PSNR = 26.18 dB
and in Figure 4(b) PSNR = 26.15 dB but a perceptual metrics like WSNR [ 8 ] , for example, as-
sesses that it is equal to 34.08 dB. Therefore, the recovered image forward quantized by pSQ
is perceptually beter than the one only quantized by an SQ. Since the later produces more
compression artifacts, the pSQ result at 0.45 bpp ( Figure 4(b) ) contains less artifacts than SQ at
0.59 bpp. For example, the Baboon 's eye is softer and beter deined using F-pSQ and it addi-
tionally saves 4.48 kB of information.