Measured honestly: KARC vs. four codecs
KARC measurably beats the classic standards JPEG and JPEG 2000. Against the modern codecs WebP and AVIF it is clearly behind — and the numbers below show exactly that, without sugar-coating.
BD-rate against all four opponents
Negative BD-rate values mean KARC needs fewer bits for the same quality (better). Positive values mean the opposite (worse).
| Codec | BD-rate PSNR | BD-rate SSIM | KARC wins | Verdict |
|---|---|---|---|---|
| JPEG (optimize=True) | −2,1 % | +9,9 % | 18/20 | KARC beats JPEG (PSNR) |
| JPEG 2000 (OpenJPEG) | −32,3 % | −45,9 % | 20/20 | KARC beats JPEG 2000 |
| WebP | +56,8 % | — | 0/20 | KARC clearly behind |
| AVIF | +90,3 % | — | 0/20 | KARC clearly behind |
JPEG 2000 was measured with the default OpenJPEG setting via Pillow (rate-based, without further tuning). A more carefully configured JPEG 2000 implementation could close the gap.
BD-rate in brief
- BD-rate (Bjøntegaard delta rate). Measures how many more or fewer bits a codec needs on average to reach the same quality — averaged over an entire quality curve, not just a single point.
- Negative = better. A negative value (e.g. −2.1%) means KARC reaches the same quality with fewer bits. A positive value (e.g. +56.8%) means KARC needs more bits — i.e. worse.
- PSNR and SSIM. PSNR measures raw signal error, SSIM measures perceived structural similarity. We report both separately because a codec can lead on one metric and trail on the other — like KARC vs. JPEG.
Methodology
20 Kodak images
The classic Kodak set of 20 photos serves as the test corpus — an established standard for comparing image codecs.
Pillow baselines
JPEG, JPEG 2000, WebP and AVIF are produced via Pillow — the same freely available library, so the comparison stays reproducible.
qstep sweep
For each image a full rate-distortion curve is generated by sweeping the quantization strength (qstep) — from high to low quality.
PSNR & SSIM
On each curve, PSNR and SSIM are measured against bitrate; from these the BD-rate per metric and image is derived — averaged over all 20 images.
Strong vs. the standards — behind the modern codecs
KARC reproducibly beats JPEG (on PSNR) and JPEG 2000 (on PSNR and SSIM) on the Kodak set. Against WebP and AVIF, however, KARC loses on every single image — the gap is large and we do not dispute it. KARC is thus an honestly competitive codec against the classic standards, but no replacement for modern methods.
Documents as PDF
The full technical documentation and the mathematical test report — both with a subtle watermark.