legal-ocr

Text extraction from scanned legal PDFs uses optical character recognition (OCR) to convert image-based documents into editable text. This Skill applies multi-engine OCR optimized for Portuguese legal language, preprocessing techniques like deskew and noise removal, and confidence scoring to produce high-accuracy output suitable for archival digitization and searchable databases.

Yes. This Skill handles multi-column layouts, tables, and variable image quality common in Brazilian legal documents—including historic archives. It automatically detects document structure (relatório, fundamentação, dispositivo sections), applies PaddleOCR with EasyOCR fallback, and flags low-confidence regions for human review.

OCR-to-text conversion with structure detection and quality validation produces embeddings suitable for RAG pipelines. This Skill extracts text while identifying document sections and applying dictionary-based corrections tuned for Portuguese legal terminology, enabling downstream semantic search and jurisprudence indexing.

Quality validation with confidence scoring and review flags enables human-in-the-loop checks, but is not required. Difficult scans automatically fall back to secondary OCR engines, and dictionary-based corrections improve accuracy on Portuguese legal language before output.

This Skill processes scanned PDFs (sentenças, petições, acórdãos) as input. Output formats support structured text extraction for use in searchable databases, RAG systems, and embeddings generation for semantic search applications.

Historic archives often have poor image quality, fading, stains, and variable paper conditions. This Skill applies CLAHE contrast enhancement, noise removal, and deskewing before dual-engine OCR to handle these cases, with confidence scoring to identify remaining problem areas.