Robust ASIC-Based Image Authentication Using Reed-Solomon LSB Watermarking

Francisco Angulo de Lafuente
Independent AI Researcher, Madrid, Spain · 2026
Winner of the NVIDIA & LlamaIndex 2024 Developer Contest. Research Interest Score 471.5, 1 citation, h-index 1.
Root CID: bafybeidhqsg5n72vhjaz2hqkj7vbwo5pwndlz24pmtbg247hjxfessgosq
A hardware-bound digital image authentication system combining ASIC proof-of-work, Reed-Solomon error correction over , and LSB steganography. Signatures survive up to 40% pixel destruction while maintaining full recovery. The formalization covers the underlying coding theory: Hamming codes (including the complete Hamming(7,4) with single-error correction), insertion/deletion codes with Levenshtein distance, and Varshamov–Tenengolts codes for single-deletion correction.
10
modules
29
theorems
731
lines
0
sorry
27
Rust tests
Formal Verification Certificate
Every theorem in this project has been machine-checked by the Lean 4 kernel. No axiom is assumed without proof. No gap exists in the verification chain.
Robust ASIC Watermarking • Lean 4 + Mathlib • Apoth3osis Labs
Paper ↔ Proof Correspondence
| PAPER | LEAN MODULE | STATUS | NOTE |
|---|---|---|---|
| Eq 2 — RS error correction capacity | Hamming.Basic, Hamming74 | FORMALIZED | Hamming codes formalized as foundation; RS codes remain a gap |
| Eq 3 — GF(2^8) field arithmetic | Hamming.Basic (GF(2)) | PARTIAL | GF(2) = ZMod 2 formalized; GF(2^8) extension field is a gap |
| Eq 4 — Log/antilog multiplication | (gap) | GAP | Requires GF(2^8) formalization with polynomial arithmetic |
| Eq 5 — LSB embedding | InsDel.Operations | PARTIAL | Bit-level insert/delete operations formalized; LSB steganography specific to images |
| Table 4 — Damage tolerance | InsDel.Codes | PARTIAL | Code predicates formalized; binomial redundancy voting is a gap |
| PoW — SHA-256 double hash | HeytingLean.Util.SHA256 | FORMALIZED | SHA-256 bytes/hex digest formalized in utility module |
| §3 — VT code error correction | VT.Basic, VT.Decode | FORMALIZED | VT codes fully formalized: membership, moment, specification-level decoder |
| §4 — Distance metrics | InsDel.Distance, InsDel.Levenshtein | FORMALIZED | Both insertion/deletion distance and LCS-based Levenshtein distance |
Headline Theorems
For any 4-bit message and any single-bit error position, syndrome decoding recovers the original codeword. Verified over all cases via native_decide.
Inserting the deleted element at the same position recovers the original. The fundamental operation underlying error correction for synchronization channels.
If a received word came from a single deletion of a unique VT codeword, the specification-level decoder recovers the original. VT codes are the asymptotically optimal family for single-deletion correction.
Proof Blueprint
“Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.”
Frank Herbert, Dune
A janitor who proves a theorem outranks a tenured professor who publishes noise.
Not as a slogan. As a structural fact of how the network operates. The only currency that matters is the quality of your contribution, measured not by committee but by mathematics.
A valid, original framing or conjecture
Formal argument with paper-level rigor
Connecting theory to observable outcomes
Working software the project depends on
Reproducible research with methodology and data
Machine-verified claim checked by a proof assistant
Foundational, load-bearing implementation
Connecting subsystems or knowledge domains end-to-end
Strategic direction & roadmaps
Writing, documentation & papers
Visual, UX & information design
Teaching, tutorials & workshops
Community, outreach & events
Partnerships, governance & policy
Translation, media & accessibility
Ethics, review & quality assurance
Every accepted contribution receives a MENTAT Contribution Record — cryptographically signed, IPFS-pinned, permanently yours. No committee decides your worth. The type checker does.
Module Dependencies
Hover over a module to trace its imports. Three independent clusters: Hamming error correction, insertion-deletion distance theory, and VT codes.
Verified C Artifacts
Compiled with gcc -std=c11 -Wall -Wextra -Werror — 0 warnings.
Safe Rust Artifacts
cargo build 0 warnings · cargo test 27/27 pass · No unsafe.
Provenance Chain
Research Paper
Angulo de Lafuente (2026). ASIC-based image authentication with RS codes over GF(2^8) and LSB steganography.
Proof Obligations
8 key claims extracted: GF arithmetic, RS error bounds, Hamming distance, VT codes, PoW hash binding, damage tolerance.
Lean 4 Formalization
10 modules, 731 lines, 30 theorems. Zero sorry/admit. lake build --wfail passes. guard_no_sorry.sh clean.
Adversarial Audit
Hostile review passed: all proofs verified, native_decide usage justified for finite exhaustive cases.
C Transpilation
10 C files compiled with gcc -std=c11 -Wall -Wextra -Werror. Zero warnings. Faithful translation of computational content.
Rust Transpilation
11 Rust files (9 modules + lib.rs + Cargo.toml). cargo build 0 warnings, cargo test 27/27 pass. No unsafe.
IPFS Pinned
All artifacts content-addressed and pinned to IPFS. CID: bafybeidhqsg5n72vhjaz2hqkj7vbwo5pwndlz24pmtbg247hjxfessgosq. Immutable, globally retrievable via any IPFS gateway.
Remaining Gaps
The following aspects of the paper are not yet formalized and represent future work:
GF(2^8) Extension Field
Splitting field of x^8 + x^4 + x^3 + x^2 + 1 over GF(2). Log/antilog multiplication tables.
Reed-Solomon Codes
RS(n,k) encoding, generator polynomial, syndrome decoding. Singleton bound d = n-k+1.
RS Error Correction Bound
t = (n-k)/2. RS(n, n-32) corrects exactly 16 symbol errors.
Redundancy Voting (Binomial)
k-of-n voting with binomial probability. Textbook result.
