ocr-pipeline

Ocr Pipeline

An invoice processing system receives scanned documents as image files. Each image needs preprocessing (deskew, contrast enhancement), OCR text extraction, field parsing (invoice number, date, line items, total), and validation that the extracted total matches the sum of line items.

Pipeline

[oc_preprocess_image]
     |
     v
[oc_extract_text]
     |
     v
[oc_validate_text]
     |
     v
[oc_structure_output]

Workflow inputs: imageUrl, documentType, language

Workers

ExtractTextWorker (task: oc_extract_text)

Extracts text from a preprocessed image using OCR.

• Writes rawText, confidence, characterCount

PreprocessImageWorker (task: oc_preprocess_image)

Preprocesses an image for OCR: deskew, binarize, contrast-enhance.

• Writes processedImage, deskewAngle, enhancementsApplied

StructureOutputWorker (task: oc_structure_output)

Organizes validated fields into a structured document.

• Writes structured, fieldCount

ValidateTextWorker (task: oc_validate_text)

Validates extracted OCR text and identifies structured fields.

• Trims whitespace
• Writes cleanText, fields, validationScore

---

17 tests | Workflow: ocr_pipeline | Timeout: 60s

See RUNNING.md for setup and usage.