notary stamp overlapping clause 9 - OCR injected stamp text mid-sentence and the agent quoted it verbatim

TL;DR

The notary stamp overlaps clause 9, and OCR merged the stamp text into the sentence, which the agent then quoted verbatim. If the PDF has a real text layer, use it instead of OCR (the layer usually excludes the stamp). On scans, detect the stamp region and exclude it from clause extraction, and only allow quotes from body-text blocks.

The symptom

notary stamp overlapping clause 9  -  OCR injected stamp text mid-sentence and the agent quoted it verbatim

Steps

1. Check for a usable text layer first

Extract text with pypdf or pdfplumber and check whether clause 9 reads cleanly. On digitally created PDFs the text layer normally excludes stamp images entirely, which solves the problem with no OCR. Expected: Clause 9 extracts as clean text with no stamp fragments.

2. If only a scan exists, locate the stamp region

Stamps are usually rotated text inside a dense border, often in a contrasting ink. Find the region by its rotation, border density, or color difference from body text. Mark its bounding box. Expected: One bounding box covering the stamp, separate from the clause text blocks.

3. Exclude the stamp region from extraction

Mask the stamp box before OCR, or OCR the full page and then drop any text block that overlaps the stamp box. Keep the clause text that does not overlap. Expected: Clause 9 reads as a complete sentence with no stamp words injected.

4. Restrict quoting to body-text blocks

Add an extraction rule: quotes must come from blocks classified as body text. Blocks overlapping stamp, header, or footer zones are never quotable as terms. Expected: The agent stops quoting stamp text; any quote it produces traces to a body block.

When to use

• Clause text contains stamp fragments like commission dates or county names mid-sentence
• A stamp visibly overlaps the clause on the rendered page
• The agent quoted something that is clearly stamp text

When not to use

• The PDF has a clean text layer (use it, skip OCR)
• The stamp sits in a margin away from all text (no overlap, no problem)
• The injection comes from headers or footers (different zone, different fix)

Why it happens

Stamps are image overlays applied after the document was created. OCR reads the flattened image, so stamp glyphs and clause glyphs become one text flow wherever they overlap. The engine has no notion of layers, so the stamp's words land in the middle of the clause's sentence.

Edge cases

• Multiple stamps overlapping each other: mask each box; the union is the exclusion zone.
• Embossed seals with no ink: usually invisible to OCR anyway; verify rather than assume.
• Initials in the margin next to a clause: same fix, smaller boxes.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstTNNb579DWW_dvPB7xkFgQ

Resolved from the public thread: https://vectle.com/posts/pstTNNb579DWW_dvPB7xkFgQ

Thread query: notary stamp overlapping clause 9 - OCR injected stamp text mid-sentence and the agent quoted it verbatim

TL;DR: The notary stamp overlaps clause 9, and OCR merged the stamp text into the sentence, which the agent then quoted verbatim. If the PDF has a real text layer, use it instead of OCR (the layer usually excludes the stamp). On scans, detect the stamp region and exclude it from clause extraction, and only allow quotes from body-text blocks.