What each exhibit claims, and what it does not.

Published technique. Longdan Tan, Yuliang Lu, Xuehu Yan, Lintao Liu, Xuan Zhou, “XOR-ed visual secret sharing scheme with robust and meaningful shadows based on QR codes,” Multimedia Tools and Applications 79 (2020), 5719–5741. doi:10.1007/s11042-019-08351-0 FULL

Detection source. Jinrui Chen, Kai Chen, Yiming Wang, Xuehu Yan, Lintao Li, “A General Steganalysis Method of QR Codes,” ICDF2C 2022, LNICST 508 (2023), 472–483. doi:10.1007/978-3-031-36574-4_28 FULL

HideAndSeen

Replaces pad codewords after the terminator, recomputes parity, decodes the overt text, recovers the secondary bytes, and checks the canonical pad pattern.

Simplified

The single-message padding exhibit is not Tan et al.'s threshold visual-sharing protocol.

Detection lesson

The prescribed 0xEC / 0x11 sequence makes this channel easy to flag.

Katarzyna Koptyra and Marek R. Ogiela, “Steganography in QR Codes—Information Hiding with Suboptimal Segmentation,” Electronics 13(13):2658 (2024). doi:10.3390/electronics13132658 META

HideAndSeen

Builds four valid numeric, alphanumeric, and byte segment sequences and maps a two-bit signal to the selected representation.

Simplified

The selection rules and optimizer are HideAndSeen's own transparent educational construction, not a byte-for-byte paper reproduction.

Detection lesson

Mode sequences and framing overhead are inspectable even when all versions decode to the same text.

Song Wan, Yuliang Lu, Xuehu Yan, Wanmeng Ding, Hanlin Liu, “High Capacity Embedding Methods of QR Code Error Correction,” LNICST 214 (2017/2018), 70–79. doi:10.1007/978-3-319-72998-5_8 FULL

Y.-J. Chiang, P.-Y. Lin, R.-Z. Wang, Y.-H. Chen, “Blind QR code steganographic approach based upon error correction capability,” KSII TIIS 7 (2013), 2527–2543. doi:10.3837/tiis.2013.10.012 BIB

HideAndSeen

Flips modules in distinct interleaved codewords, compares spread and concentrated errors at a fixed count, tracks every per-block correction limit, and asks ZXing to decode the altered matrix.

Simplified

The lower-right data-region option follows Wan's spatial rule but does not implement the paper's maximum rectangle calculation across all 40 versions.

Detection lesson

Correction is block-local: the same global error count can survive when spread and fail when concentrated.

Katarzyna Koptyra and Marek R. Ogiela, “Multi-secret Steganography in QR Codes,” WSEAS Transactions on Information Science and Applications 21 (2024), 533–537. doi:10.37394/23209.2024.21.49 META

HideAndSeen

Combines a segment-selection signal with controlled correctable module changes around one overt message.

Simplified

Uses this project's transparent channel rules rather than reproducing the paper's exact embeddings.

Detection lesson

Different readers can expose different layers, and each layer leaves different evidence.

Iuliia Tkachenko, William Puech, Christophe Destruel, Olivier Strauss, Jean-Marc Gaudin, Christian Guichard, “Two-Level QR Code for Private Message Sharing and Document Authentication,” IEEE TIFS 11(3) (2016), 571–583. doi:10.1109/TIFS.2015.2506546 BIB

HideAndSeen

Places equal-area orientation marks inside dark data modules and simulates magnification, blur, and detail loss.

Simplified

Does not reproduce the paper's textured patterns, authentication process, or print-copy response.

Detection lesson

High-frequency structure can disappear under ordinary sampling yet remain measurable in the source image.

Tan et al., “XOR-ed visual secret sharing scheme with robust and meaningful shadows based on QR codes,” Multimedia Tools and Applications 79 (2020), 5719–5741. doi:10.1007/s11042-019-08351-0 FULL

HideAndSeen

Creates two random-grid XOR shadows, embeds each bitstream in one QR's padding, independently decodes both overt payloads, extracts both shares, and XORs them to restore the image.

Simplified

Implements the paper's 2-of-2 mechanism for a small browser-generated image; it does not reproduce the full k-of-n construction or physical attack campaign.

Detection lesson

Each QR's padding is anomalous even though neither shadow alone reveals the secret image.

Kuo-Cheng Chou and Ran-Zan Wang, “Dual-Message QR Codes,” Sensors 24(10):3055 (2024). doi:10.3390/s24103055 FULL

Earlier lineage: Guo-Jian Chou and Ran-Zan Wang, “The Nested QR Code,” IEEE Signal Processing Letters 27 (2020), 1230–1234. doi:10.1109/LSP.2020.3006375 BIB

HideAndSeen

Combines equal-version QR matrices into four near/far two-state module blocks: the centered region carries the near bit and the outer area carries the far bit.

Simplified

Browser resampling is deterministic; physical phone behavior still depends on display density, focus, reader, and distance.

Detection lesson

One printed image can yield different matrices under different spatial sampling conditions.

Jinrui Chen, Kai Chen, Yiming Wang, Xuehu Yan, Lintao Li, “A General Steganalysis Method of QR Codes,” ICDF2C 2022, LNICST 508 (2023), 472–483. doi:10.1007/978-3-031-36574-4_28 FULL

HideAndSeen

Decodes, regenerates, and compares a clean symbol, a padding-channel symbol, and an innocent symbol from another encoder.

Simplified

The exhibit does not claim to reproduce every filtering stage or evaluated spatial scheme in the paper.

Detection lesson

Reference-encoder mismatch can create differences that are not hidden data.

Primary evidence is produced by this repository's five-encoder harness. A related mature methodology is Yongjian Kee, Michael K. Johnson, Hany Farid, “Digital Image Authentication from JPEG Headers,” IEEE TIFS 6(3) (2011). DATA

HideAndSeen

Compares live HideAndSeen and node-qrcode output, then summarizes held-out profiles from segno, python-qrcode, qrcodegen, libqrencode, and node-qrcode.

Simplified

The five libraries and their defaults are a controlled sample, not the population of QR generators.

Detection lesson

Implementation traces overlap and can be configured or imitated; they are not identities.

The result is generated by src/attribute.py after fixing complete segment-stream parsing. Methodological context: Andrew D. Ker et al., “Moving Steganography and Steganalysis from the Laboratory into the Real World,” IH&MMSec 2013. DATA

HideAndSeen

A random forest trained on 2,500 symbols and tested on 1,250 symbols from disjoint payloads reaches 53.52% five-way accuracy versus 20% chance, with zero skipped symbols.

Simplified

Five implementations, synthetic payload classes, and default configurations do not support broad forensic generalization.

Detection lesson

Better-than-chance classification is scientifically interesting but insufficient for strong attribution.

Phil Sallee, “Model-Based Steganography,” IWDW 2003, LNCS 2939. doi:10.1007/978-3-540-24624-4_12 2ND

Maciej Liśkiewicz, Rüdiger Reischuk, Ulrich Wölfel, “Grey-box Steganography,” Theoretical Computer Science 505 (2013). doi:10.1016/j.tcs.2012.06.005 2ND

HideAndSeen

Compares categorical cover samples, direct uniform secret mapping, and CDF-shaped selection with a simple chi-square statistic.

Simplified

No named steganographic algorithm, security proof, or operational channel is reproduced.

Detection lesson

Passing one marginal-distribution test does not establish steganographic security.

This capstone combines the implemented mechanisms and analytical checks above rather than reproducing one paper. DATA

HideAndSeen

Randomizes clean, padding, segmentation, correctable-error, and alternate-encoder samples and exposes eight analyst tools.

Simplified

The cases are synthetic and limited to mechanisms implemented by this site.

Detection lesson

There is no universal steganography bit; evidence must be interpreted against the correct structural layer and model of innocent variation.

The exhibit is Bayes' rule applied to the detectors built above, and can be checked by hand from its own grid. Motivation specific to this field: Andrew D. Ker, Patrick Bas, Rainer Böhme, Rémi Cogranne, Scott Craver, Tomáš Filler, Jessica Fridrich, Tomáš Pevný, “Moving Steganography and Steganalysis from the Laboratory into the Real World,” IH&MMSec 2013, 45–58. doi:10.1145/2482513.2482965 2ND

HideAndSeen

Varies prevalence, sensitivity and specificity over a fixed population of 10,000 symbols and reports the positive predictive value, with one dot per symbol.

Simplified

A single detector applied to an independent population. Real analysis involves correlated traffic, multiple instruments, and analyst judgement between them.

Detection lesson

Sensitivity and specificity describe the instrument; only prevalence determines what one of its alarms is worth. When hiding is rare, specificity dominates.

The coverage map is this repository's own mechanisms and checks, and can be verified against the exhibits it names. The wider argument that laboratory detection rates overstate real-world performance is Andrew D. Ker, Patrick Bas, Rainer Böhme, Rémi Cogranne, Scott Craver, Tomáš Filler, Jessica Fridrich, Tomáš Pevný, “Moving Steganography and Steganalysis from the Laboratory into the Real World,” IH&MMSec 2013, 45–58. doi:10.1145/2482513.2482965 2ND

HideAndSeen

Maps the five hiding channels implemented here against the six checks demonstrated here, then reports which channels an informed adversary could still use, and what the selected checks cost in false alarms at a 1-in-1,000 base rate.

Simplified

No adaptive-adversary experiment is run. Checks are treated as independent, which is a simplification rather than a bound: it overstates combined coverage and combined alarm volume together, and because false alarms dominate precision at a 1-in-1,000 base rate, its net effect there is unsigned. The per-check error rate is a parameter the reader sets rather than a measured property.

Detection lesson

A detection rate measured against samples not built to evade it is an upper bound. Closing a coverage gap costs false alarms, so complete coverage against a rare target can produce a queue in which almost nothing is real.