New lossy contour coding slashes bitrates by 66% using decoder-side guidance
Coarse-to-fine anchor refinement achieves up to 5x compression over JBIG2 with high geometric accuracy.
Deep Dive
The paper proposes a decoder-guided lossy contour coding framework that models high-quality contours as refinements of coarse contours already available at the decoder. By adaptively skipping anchors under distortion constraints, the method achieves 54.5%–66.9% bitrate reduction over non-guided methods and up to 5× savings over JBIG2, enabling efficient side-information transmission for vision tasks over bandwidth-limited channels.
Key Points
- Achieves 54.5%–66.9% bitrate reduction compared to contour coding without decoder-side guidance.
- Up to 5× compression savings over the standard JBIG2 codec while maintaining geometric accuracy.
- Uses adaptive anchor skipping with explicit rate-distortion control, leveraging coarse contours already available at the decoder.
Why It Matters
Enables high-quality structural side information for vision tasks over extremely bandwidth-limited channels, improving edge-AI and remote sensing applications.