feat: OCR pipeline and context optimization - #8
Merged
Conversation
- Add scholardoc/ocr/ module with 6 files (dictionary, linebreak, detector, reocr, pipeline, __init__) - Update pyproject.toml with two-tier OCR strategy: - [ocr]: CPU-optimized with Tesseract (~1.35s per dirty page) - [ocr-gpu]: GPU-optimized with docTR + Tesseract fallback (~0.45s per dirty page) - Core dependencies: pymupdf, pyspellchecker, pillow only - OCR engines now optional for maximum compatibility Part of Phase 1 milestone 1.6 (OCR correction pipeline) Implements plan from .claude/logs/plan_ocr_integration.md
Moves 5 large plan files (~8k tokens) to .claude/logs/archive/ to reduce context overhead during sessions. Archived: - plan_ocr_implementation_session2.md - plan_ocr_implementation_session2_REVISED.md - plan_ocr_integration.md - plan_convert_orchestrator.md - plan_document_profiles.md Session logs retained for /project:analyze-logs
- CLAUDE.md: 523 lines → 68 lines (87% reduction, ~4k token savings) - Extracted to docs/: - VISION.md - Full vision and applications - TESTING_METHODOLOGY.md - Validation guidelines - COMMANDS.md - Full command reference - RULES.md - Development rules - GIT_WORKFLOW.md - Branching strategy Part of context optimization initiative.
Phase 4-5 of context optimization: - Extract automation examples from analyze-logs.md to docs/ - Extract templates from init.md to docs/INIT_TEMPLATES.md - Condense output templates in improve.md and review-pr.md - Remove unused Context7 MCP from permissions Total word reduction: ~1327 words across 4 commands
- Move 5 commands from .claude/commands/ to .claude/agents/ - analyze-logs, create-pr, diagnose, improve, review-pr - Add session logs for development tracking - Add backup files to .gitignore
Production-ready OCR correction pipeline for scholarly PDF extraction. Components added to scholardoc/ocr/: - AdaptiveDictionary: Morphological word validation with scholarly vocabulary - OCRErrorDetector: Pattern-based error detection (96.9% detection rate) - LineBreakRejoiner: Block-based hyphenation handling (ADR-003) - HybridReOCREngine: 4-tier fallback (docTR GPU → Tesseract → docTR CPU → skip) - OCRPipeline: Main orchestrator coordinating all components Integration: - Add OCRConfig to config.py (disabled by default for backward compatibility) - Add OCRErrorType, OCRCorrectionRecord enums to models.py - Integrate conditional OCR processing in convert.py Validation results: - Detection rate: 96.9% on 130 error pairs - False positive rate: 20.8% (target ≤25%) - 320 tests passing, including 37 new OCR unit tests Closes Phase 1 of the OCR pipeline roadmap.
- Add 16 multilingual edge case tests: - Accented character handling (French, German) - Case sensitivity for scholarly vocabulary - Latin phrase recognition - Greek transliterations - Mixed-language text - OCR corruption scenarios - Edge cases (empty strings, punctuation, hyphenated terms) - Add 7 validation set integration tests: - Detection rate verification (>= 95% target) - Detection rate by error type - False positive rate verification (<= 25% target) - Scholarly vocabulary verification - Pipeline integration with validation data - Regression tests for known difficult cases - Add 6 performance tests: - Dictionary lookup speed (< 1ms per word) - Morphology check speed (< 5ms per word) - Detector page speed (< 50ms for 500 words) - Pipeline page speed (< 100ms for 500 words) - Bulk processing performance - Memory leak detection Total: 349 tests passing (up from 320)
- Add OCRSourceInfo dataclass to models.py for tracking OCR engine metadata - Parse PDF producer/creator fields to identify OCR engine (Adobe, ABBYY, Tesseract) - Integrate OCR source extraction into DocumentBuilder.build() - Store in QualityInfo.ocr_source for traceability Documentation: - Add ADR-004: OCR Source Tracking and Engine Validation - Document current validation coverage (Adobe Paper Capture 2009-2023) - Add multi-OCR-engine validation to future considerations in ROADMAP.md Analysis: Current samples use Adobe Paper Capture only (single vendor, 14-year span). Detection approach is engine-agnostic (dictionary + patterns), so should generalize.
Owner
Author
Code Review: Phase 1 OCR IntegrationReviewer: Claude Opus 4.5 SummaryThis PR implements a well-designed OCR error detection and correction pipeline for scholarly documents. The architecture follows the validated ADR-002 approach of using spellcheck as a selector for re-OCR rather than auto-correction - the correct choice for scholarly text with specialized vocabulary. Metrics
Strengths
Minor Issues (Non-blocking)
Suggestions for Future
Risk Assessment
Reviewed using Opus 4.5. Full review saved to project memory. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Changes
OCR Pipeline
Context Optimization (Phases 1-5)
Test Plan
/project:init validateto verify setup