StegoBench
Analyze • Detect • Stay Ahead
Carrier & payload
Start by picking the image(s) to hide your message in, then write the secret and choose how to lock it.
Any file up to 64 KB works.
Methods & run
Choose which methods to compare and how hard to push them, then hit run. Grab a coffee if you picked a big set.
Leaderboard
Ranked by what matters to you. Change the goal and the order updates instantly.
Recommendations
Is the ranking reliable? Statistics
How the scores are calculated
Recovery = chance-adjusted bit recovery of the framed payload after the attack and error correction (50 % match = 0 %, 100 % = exact). Robustness = mean recovery over all supported attacks.
Imperceptibility = 70 % PSNR (30 dB scores 0, 60 dB scores 100) + 30 % SSIM (11×11 Gaussian window; 0.95 scores 0, 1.00 scores 100).
Stealth = 100 − detection risk. With fewer than 12 image-key samples the risk is a heuristic: 30 % visible distortion, 50 % strongest of RS / Weighted Stego / progressive chi-square (each measured against the cover), 20 % growth of high-frequency noise. With 12 or more samples it comes from the measured ROC area of the best classical detector: stealth = 100 × (1 − max(0, 2·AUC − 1)).
Capacity = log-scaled maximum payload. Overall = weighted sum with the weights of the selected goal. “#1 in x %” re-ranks 600 random weightings to show how much the winner depends on the weights you chose.
RS, WS and chi-square target LSB replacement. They are classical detectors, not trained steganalysers (SRM, SRNet): treat “undetected” as “not caught by these tests”. See the README for cross-checking with Aletheia.
Reproducibility card
Deep analysis
The bigger questions: what happens as the message grows, where each method finally breaks, and what a detector actually sees.
Embeds random payloads of increasing size (bits per pixel) with the methods ticked in step 2, then applies one attack. Uses up to three of your cover images, no encryption or ECC, so it shows the raw channel.
Recovery after the attack
Image quality (PSNR)
Applies one attack at increasing strength to a fixed 24-byte payload and plots how much of it survives. The drop-off point is the method’s breaking point.
ROC curves (best classical detector per method)
Extract & verify
Got a stego image you saved earlier? Load it here and read the hidden message back. The checksum tells you whether it worked.
Extracted text will appear here.