OCR & Entity Extraction Benchmarks

Test Case

Image: harry-potter-wiki-frontpage.png — a Wikipedia screenshot containing mixed text, headers, tables, and image captions.

Pipeline Metrics

Metric	Result
Upload	PASS
Upload time	0.01s
Processing time	8.07s
Total latency	8.08s
OCR detection	PASS
Searchable text	PASS
Chunks created	2

OCR Metadata

The system automatically triggers OCR for images (PNG/JPG/JPEG) and PDF pages with very little native text (<80 characters). The metadata.quality block confirmed OCR was applied:

{
  "quality": {
    "text_extraction": "ocr",
    "ocr_used": true,
    "warnings": [
      {
        "code": "ocr_applied",
        "message": "Text was extracted with OCR because no native text was available."
      }
    ]
  }
}

Search Retrieval

Query: "Harry Potter is a series of seven fantasy novels" (explicit Content scope)

Metric	Value
Top Document Version	harry-potter-wiki-frontpage.png (unversioned)
Top rank	1
Content evidence items	1
Query time	14ms
Category	other

The caller-facing response is an Evidence Result page. Its totals and pagination count Document Versions rather than chunks; OCR text and metadata remain untrusted evidence.

Entity Extraction (spaCy NER)

The spaCy NER pipeline extracted named entities from OCR-processed text:

Entity Type	Extracted
PERSON	Harry Potteris, J. K.Rowling, Ron Hary Potter Weasley, Hermione Granger, knownin-universeasMuggles
ORG	Wikipedia, Hogwarts School of Witchcraft andWizardry, the Ministry of Magic, Philosopher's Stone, Chamber of Secrets, Azkaban, Scholastic Press, Deathly Hallows
DATE	1997, 1998, 1999, 2000, 2003, 2007
LOCATION	the United States

OCR Text Quality Notes

OCR artifacts expected

RapidOCR produced artifacts on this complex Wikipedia layout. Examples from the benchmark output:

• HarryPotter (missing space)
• Harry Potteris (merged phrase)
• knownin-universeasMuggles (merged words)

These artifacts are expected for complex image layouts and did not prevent effective semantic search.

Classification

The live search result classified the non-financial OCR content as Category: other.