Skip to content

scanned_pdfs_and_quality.md

Latest commit

 

History

History
174 lines (123 loc) · 6.59 KB

scanned_pdfs_and_quality.md

File metadata and controls

174 lines (123 loc) · 6.59 KB

Scanned PDFs and Quality: OCR Parsing Guardrails

🧭 Quick Return to Map

You are in a sub-page of OCR_Parsing.
To reorient, go back here:

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

Stabilize OCR extraction on noisy scans, low-resolution images, and multi-generation photocopies. Ensure text is auditable, retrievable, and bound by schema despite quality issues.

Open these first

Acceptance targets

  • OCR character error rate (CER) ≤ 2% after cleanup
  • ΔS(question, retrieved) ≤ 0.45 even when scan quality < 300 dpi
  • λ remains convergent across paraphrases
  • All extracted text auditable against source image hash

Typical failure signatures → fix

  • Broken characters and merged glyphs
    Apply normalization and Unicode repair before indexing. Validate against a whitelist of expected ranges.

  • Multi-generation photocopy blur
    Route through OCR engine supporting adaptive binarization. Anchor outputs with image hash to avoid ghost drift.

  • Double-encoded PDFs (text + image overlay)
    Deduplicate layers. Choose the higher-confidence text layer and tag source.

  • Skewed pages or rotated scans
    Run deskew filter before OCR. Capture skew angle metadata for audit.

  • Mixed-language or font variants
    Force language models per region. Split by script. Store per-block language code.

  • Noise artifacts (staple marks, stamps, watermarks)
    Strip bounding boxes below token threshold. Mark as noise_block instead of narrative text.

Fix in 60 seconds

  1. Hash source image
    Store scan_id and image_hash for every page. Tie all extracted text back to this anchor.

  2. Normalize text
    Apply Unicode NFKC. Collapse broken ligatures and fix spacing errors.

  3. De-layer double PDFs
    Choose the OCR text layer with confidence ≥ 0.90. Drop shadow text.

  4. Audit with ΔS
    Probe scanned text with 3 paraphrases. If ΔS ≥ 0.60, run re-OCR with stricter binarization.

  5. Chunk and contract
    Split by page. Enforce data contract fields: page_no, scan_id, text_clean, bbox.

Minimal recipes by engine

  • Google Document AI
    Use qualityScores.confidence field. Reject blocks with confidence < 0.7.

  • AWS Textract
    Hash BlockType=PAGE. Keep page-level confidence. Store as scan_id.

  • Azure OCR
    Normalize boundingRegions. Add language code explicitly if detected.

  • ABBYY
    Use <charParams> confidence. Flag low confidence segments for secondary OCR.

  • PaddleOCR
    Use angle classification for deskew. Split multilingual pages into per-line language tags.

Data contract extension


{
"scan\_id": "p12\_imghash",
"page\_no": 12,
"image\_hash": "sha256:...",
"text\_clean": "...",
"language": "en",
"confidence": 0.92,
"noise\_blocks": \[...],
"source\_url": "..."
}

Verification

  • Leak check: ensure no shadow/duplicate text.
  • Quality probe: CER ≤ 2% on 1k sample chars.
  • Stability probe: ΔS stable across paraphrases.
  • Auditability: all text traceable to image hash.

Copy-paste LLM prompt


You have TXTOS and WFGY Problem Map.

My scan:

* page\_no: {n}
* text\_clean: "..."
* confidence: 0.xx
* image\_hash: "..."

Tasks:

1. If text looks corrupted, fail fast and cite fix page.
2. Validate schema (ocr-parsing-checklist, data-contracts).
3. Return JSON: { "answer":"...", "citations":\[...], "ΔS":0.xx, "λ\_state":"..." }

🔗 Quick-Start Downloads (60 sec)

Tool Link 3-Step Setup
WFGY 1.0 PDF Engine Paper 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>”
TXT OS (plain-text OS) TXTOS.txt 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

Explore More

Layer Page What it’s for
⭐ Proof WFGY Recognition Map External citations, integrations, and ecosystem proof
⚙️ Engine WFGY 1.0 Original PDF tension engine and early logic sketch (legacy reference)
⚙️ Engine WFGY 2.0 Production tension kernel for RAG and agent systems
⚙️ Engine WFGY 3.0 TXT based Singularity tension engine (131 S class set)
🗺️ Map Problem Map 1.0 Flagship 16 problem RAG failure taxonomy and fix map
🗺️ Map Problem Map 2.0 Global Debug Card for RAG and agent pipeline diagnosis
🗺️ Map Problem Map 3.0 Global AI troubleshooting atlas and failure pattern map
🧰 App TXT OS .txt semantic OS with fast bootstrap
🧰 App Blah Blah Blah Abstract and paradox Q&A built on TXT OS
🧰 App Blur Blur Blur Text to image generation with semantic control
🏡 Onboarding Starter Village Guided entry point for new users

If this repository helped, starring it improves discovery so more builders can find the docs and tools.
GitHub Repo stars

要不要我直接幫你接續做下一個 multi_language_and_fonts.md？