Rate Shaping through Feature Analysis
System Overview
Figure 1: The proposed system diagram. The output bit rate of the rate
shaper R(Y) should not exceed the bit rate constraint R_{budget}, which
could vary with time.
Feature Analysis
Figure 2: The image contours of (a) at a scale is shown in (b).
The bottom row is the result of thresholding on importance(C) for every
contour C: (c) the remaining contours, and (d) the corresponding
image blocks.
Comparison of Different Approaches
Comparison of different rate shaping approaches
for ``baboon" image, pre-compressed by JPEG at 1.3 bpp.
In these graphs, the upper envelops of the solid lines are the
optimal solutions.
JPEG Example
Figure 3: Reconstructed ``baboon" images, pre-compressed
by JPEG at 1.3 bpp, and rate-shaped at
0.367 bpp. Top: block-dropping+adaptive coefficient truncation scheme,
PSNR=21.66 dB; Bottom: block-dropping+nonadaptive
coefficient truncation scheme, PSNR=21.37 dB.
MPEG Example
Figure 5: Reconstructed I frames (luminance only) of the ``Bicycle" sequence,
precompressed by MPEG at 1.0 bpp, and rate-shaped at 0.51 bpp. Top:
block-dropping+adaptive coefficient truncation scheme,
PSNR=26.87 dB; Bottom: block-dropping+nonadaptive coefficient
truncation scheme, PSNR=25.54 dB.
People:
References:
Acknowledgements:
Last update : Dec. 28, 1996.
If images on this page look dark to you, see the note about
gamma correction.
back to the main page