correction_post_processor.py

#!/usr/bin/env python3
"""
Correction Post-Processor for OCR Results
Handles text corrections and improvements based on rules
"""

import json
import re
import logging
from pathlib import Path

logger = logging.getLogger(__name__)

class CorrectionPostProcessor:
    """Post-process OCR results with correction rules"""
    
    def __init__(self, rules_file='correction_rules.json'):
        self.rules_file = rules_file
        self.rules = self.load_rules()
        self.stats = {
            'total_processed': 0,
            'total_corrected': 0,
            'text_fixed': 0,
            'box_moved': 0,
            'new_zone': 0,
            'deleted': 0,
            'validated': 0
        }
    
    def load_rules(self):
        """Load correction rules from JSON file"""
        try:
            if Path(self.rules_file).exists():
                with open(self.rules_file, 'r', encoding='utf-8') as f:
                    rules = json.load(f)
                logger.info(f"Loaded {len(rules.get('text_replacements', {}))} correction rules")
                return rules
            else:
                logger.warning(f"Rules file {self.rules_file} not found, using default rules")
                return self.get_default_rules()
        except Exception as e:
            logger.error(f"Failed to load correction rules: {e}")
            return self.get_default_rules()
    
    def get_default_rules(self):
        """Get default correction rules"""
        return {
            "text_replacements": {
                # Common OCR mistakes
                "O": "0",  # Letter O to number 0 in dimensions
                "I": "1",  # Letter I to number 1 in dimensions
                "l": "1",  # lowercase l to number 1
                "S": "5",  # Letter S to number 5
                "G": "6",  # Letter G to number 6
                "B": "8",  # Letter B to number 8
                # Dimension-specific corrections
                "Ø": "Ø",  # Keep diameter symbol
                "±": "±",  # Keep plus-minus symbol
                # Remove common OCR artifacts
                "|": "",   # Remove stray vertical bars
                "~": "",   # Remove tildes
                "`": "",   # Remove backticks
            },
            "dimension_patterns": [
                r"(\d+\.?\d*)\s*±\s*(\d+\.?\d*)",  # ± tolerance
                r"(\d+\.?\d*)\s*\+\s*(\d+\.?\d*)\s*/\s*-\s*(\d+\.?\d*)",  # +tolerance/-tolerance
                r"(\d+\.?\d*)\s*-\s*(\d+\.?\d*)",  # -tolerance only
                r"(\d+\.?\d*)\s*\+\s*(\d+\.?\d*)",  # +tolerance only
            ],
            "confidence_threshold": 0.7,
            "min_text_length": 1
        }
    
    def process_zones(self, zones):
        """Process OCR zones with correction rules"""
        if not zones:
            return zones
        
        self.stats['total_processed'] = len(zones)
        corrected_zones = []
        
        for zone in zones:
            try:
                corrected_zone = self.correct_zone(zone)
                if corrected_zone:
                    corrected_zones.append(corrected_zone)
                    
                    # Track corrections
                    if self.was_corrected(zone, corrected_zone):
                        self.stats['total_corrected'] += 1
                        
            except Exception as e:
                logger.warning(f"Failed to correct zone: {e}")
                corrected_zones.append(zone)
        
        logger.info(f"Processed {self.stats['total_processed']} zones, corrected {self.stats['total_corrected']}")
        return corrected_zones
    
    def correct_zone(self, zone):
        """Apply corrections to a single zone"""
        if not zone or not zone.get('text'):
            return zone
        
        original_text = zone['text']
        corrected_text = self.correct_text(original_text)
        
        # Create corrected zone
        corrected_zone = zone.copy()
        corrected_zone['text'] = corrected_text
        corrected_zone['original_text'] = original_text
        corrected_zone['was_corrected'] = corrected_text != original_text
        
        return corrected_zone
    
    def correct_text(self, text):
        """Apply text corrections based on rules"""
        if not text:
            return text
        
        corrected = text
        
        # Apply character replacements
        replacements = self.rules.get('text_replacements', {})
        for wrong, correct in replacements.items():
            corrected = corrected.replace(wrong, correct)
        
        # Apply regex patterns for dimensions
        dimension_patterns = self.rules.get('dimension_patterns', [])
        for pattern in dimension_patterns:
            # This is just for validation, actual parsing is done elsewhere
            if re.search(pattern, corrected):
                break
        
        # Clean up whitespace
        corrected = re.sub(r'\s+', ' ', corrected).strip()
        
        return corrected
    
    def was_corrected(self, original_zone, corrected_zone):
        """Check if zone was actually corrected"""
        if not original_zone or not corrected_zone:
            return False
        
        original_text = original_zone.get('text', '')
        corrected_text = corrected_zone.get('text', '')
        
        return original_text != corrected_text
    
    def get_stats(self):
        """Get correction statistics"""
        return self.stats.copy()
    
    def reset_stats(self):
        """Reset statistics"""
        self.stats = {
            'total_processed': 0,
            'total_corrected': 0,
            'text_fixed': 0,
            'box_moved': 0,
            'new_zone': 0,
            'deleted': 0,
            'validated': 0
        }