demo: Test d'anonymisation sur document réel

- Test sur 003_simple_compte_rendu_CRO_23155084.pdf
- 25 PII détectés (4 sur page principale + propagation globale)
- Types: NOM, ADRESSE, CODE_POSTAL, DATE_NAISSANCE
- Validation: AUCUNE FUITE détectée ✓
- Scripts d'analyse: analyze_anonymization_result.py, demo_complete_anonymization.py
- Résultats dans tests/ground_truth/pdfs/anonymized_test/

analyze_anonymization_result.py

@@ -0,0 +1,122 @@
+#!/usr/bin/env python3
+"""
+Analyse des résultats d'anonymisation.
+"""
+import json
+from pathlib import Path
+from collections import Counter
+from evaluation import LeakScanner
+
+def main():
+    # Fichiers générés
+    base_name = "003_simple_compte_rendu_CRO_23155084"
+    output_dir = Path("tests/ground_truth/pdfs/anonymized_test")
+    
+    audit_path = output_dir / f"{base_name}.audit.jsonl"
+    redacted_pdf = output_dir / f"{base_name}.redacted_raster.pdf"
+    text_path = output_dir / f"{base_name}.pseudonymise.txt"
+    
+    print("="*80)
+    print("ANALYSE DES RÉSULTATS D'ANONYMISATION")
+    print("="*80)
+    print(f"\n📄 Document: {base_name}.pdf")
+    print(f"   Type: Compte-rendu opératoire (CRO)")
+    
+    # Analyser l'audit
+    if audit_path.exists():
+        print(f"\n📊 ANALYSE DE L'AUDIT")
+        print(f"   Fichier: {audit_path.name}")
+        
+        pii_list = []
+        with open(audit_path, 'r', encoding='utf-8') as f:
+            for line in f:
+                if line.strip():
+                    pii_list.append(json.loads(line))
+        
+        print(f"\n   Total PII détectés: {len(pii_list)}")
+        
+        # Compter par type
+        type_counts = Counter(pii['kind'] for pii in pii_list)
+        
+        print(f"\n   Répartition par type:")
+        for pii_type, count in sorted(type_counts.items(), key=lambda x: -x[1]):
+            print(f"     {pii_type:20s} : {count:3d}")
+        
+        # Afficher les PII uniques (page 0 uniquement)
+        page0_pii = [p for p in pii_list if p.get('page') == 0]
+        
+        if page0_pii:
+            print(f"\n   PII détectés sur la page principale:")
+            for pii in page0_pii:
+                original = pii.get('original', '')[:60]
+                print(f"     • {pii['kind']:20s} : {original}")
+        
+        # Afficher les noms extraits (propagation globale)
+        extracted_names = [p for p in pii_list if p.get('kind') == 'NOM_EXTRACTED']
+        if extracted_names:
+            unique_names = set(p['original'] for p in extracted_names)
+            print(f"\n   Noms propagés globalement ({len(unique_names)} uniques):")
+            for name in sorted(unique_names):
+                count = sum(1 for p in extracted_names if p['original'] == name)
+                print(f"     • {name:20s} : {count} occurrences")
+    
+    # Afficher le texte anonymisé
+    if text_path.exists():
+        print(f"\n📝 TEXTE ANONYMISÉ")
+        print(f"   Fichier: {text_path.name}")
+        
+        with open(text_path, 'r', encoding='utf-8') as f:
+            text = f.read()
+        
+        print(f"\n   Extrait (200 premiers caractères):")
+        print("   " + "-"*76)
+        lines = text[:200].split('\n')
+        for line in lines[:5]:
+            print(f"   {line}")
+        print("   " + "-"*76)
+    
+    # Scanner les fuites
+    if redacted_pdf.exists() and audit_path.exists():
+        print(f"\n🔒 SCAN DE FUITE")
+        print(f"   PDF anonymisé: {redacted_pdf.name}")
+        
+        scanner = LeakScanner()
+        leak_report = scanner.scan(redacted_pdf, audit_path)
+        
+        if leak_report.is_safe:
+            print(f"\n   ✓ DOCUMENT SÛR")
+            print(f"     Aucune fuite détectée")
+        else:
+            print(f"\n   ✗ ATTENTION - {leak_report.leak_count} fuite(s)")
+            
+            # Par sévérité
+            print(f"\n   Fuites par sévérité:")
+            for severity, count in sorted(leak_report.severity_counts.items()):
+                print(f"     {severity:10s} : {count}")
+            
+            # Détails
+            print(f"\n   Détails des fuites:")
+            for i, leak in enumerate(leak_report.leaks[:10], 1):
+                print(f"     {i}. [{leak['severity']}] {leak['type']}")
+                print(f"        {leak['message']}")
+            
+            if leak_report.leak_count > 10:
+                print(f"     ... et {leak_report.leak_count - 10} autres")
+    
+    print("\n" + "="*80)
+    print("✨ Analyse terminée")
+    print("="*80)
+    
+    print(f"\n💡 Fichiers disponibles:")
+    print(f"   - PDF anonymisé (raster): {redacted_pdf.name}")
+    print(f"   - PDF anonymisé (vector): {base_name}.redacted_vector.pdf")
+    print(f"   - Texte anonymisé: {text_path.name}")
+    print(f"   - Audit complet: {audit_path.name}")
+    
+    print(f"\n📂 Répertoire: {output_dir}")
+    
+    print(f"\n🔍 Pour voir le PDF:")
+    print(f"   xdg-open {redacted_pdf}")
+
+if __name__ == "__main__":
+    main()

demo_complete_anonymization.py

@@ -0,0 +1,142 @@
+#!/usr/bin/env python3
+"""
+Démonstration complète : Anonymisation + Analyse
+"""
+import json
+from pathlib import Path
+from collections import Counter
+from evaluation import LeakScanner
+
+def show_comparison():
+    """Affiche une comparaison avant/après."""
+    
+    print("\n" + "="*80)
+    print("DÉMONSTRATION COMPLÈTE : ANONYMISATION D'UN DOCUMENT RÉEL")
+    print("="*80)
+    
+    # Fichiers
+    original_pdf = Path("tests/ground_truth/pdfs/003_simple_compte_rendu_CRO_23155084.pdf")
+    output_dir = Path("tests/ground_truth/pdfs/anonymized_test")
+    base_name = "003_simple_compte_rendu_CRO_23155084"
+    
+    audit_path = output_dir / f"{base_name}.audit.jsonl"
+    redacted_pdf = output_dir / f"{base_name}.redacted_raster.pdf"
+    
+    print(f"\n📄 DOCUMENT TRAITÉ")
+    print(f"   Original: {original_pdf.name}")
+    print(f"   Type: Compte-rendu opératoire (CRO)")
+    print(f"   Complexité: Simple (1 page)")
+    
+    # Extraire le texte original
+    try:
+        import fitz
+        doc = fitz.open(original_pdf)
+        original_text = doc[0].get_text()
+        doc.close()
+        
+        print(f"\n📝 TEXTE ORIGINAL (extrait):")
+        print("   " + "-"*76)
+        lines = original_text.split('\n')[:8]
+        for line in lines:
+            if line.strip():
+                print(f"   {line[:76]}")
+        print("   " + "-"*76)
+    except Exception as e:
+        print(f"   ⚠ Impossible d'extraire le texte: {e}")
+    
+    # Analyser les PII détectés
+    if audit_path.exists():
+        print(f"\n🔍 PII DÉTECTÉS PAR LE SYSTÈME")
+        
+        pii_list = []
+        with open(audit_path, 'r', encoding='utf-8') as f:
+            for line in f:
+                if line.strip():
+                    pii_list.append(json.loads(line))
+        
+        # PII de la page principale
+        page0_pii = [p for p in pii_list if p.get('page') == 0]
+        
+        print(f"\n   Sur la page principale ({len(page0_pii)} PII):")
+        for pii in page0_pii:
+            kind = pii['kind']
+            original = pii.get('original', '')
+            print(f"     ✓ {kind:20s} : {original}")
+        
+        # Noms propagés
+        extracted = [p for p in pii_list if 'EXTRACTED' in p.get('kind', '') or 'GLOBAL' in p.get('kind', '')]
+        if extracted:
+            unique_names = set(p['original'] for p in extracted if 'NOM' in p.get('kind', ''))
+            print(f"\n   Noms propagés sur tout le document ({len(unique_names)} uniques):")
+            for name in sorted(unique_names):
+                print(f"     → {name}")
+        
+        # Statistiques
+        type_counts = Counter(pii['kind'] for pii in pii_list)
+        print(f"\n   📊 STATISTIQUES:")
+        print(f"      Total PII: {len(pii_list)}")
+        print(f"      Types différents: {len(type_counts)}")
+        
+        # Top 3
+        print(f"\n      Top 3 des types:")
+        for pii_type, count in type_counts.most_common(3):
+            print(f"        {pii_type:20s} : {count}")
+    
+    # Texte anonymisé
+    text_path = output_dir / f"{base_name}.pseudonymise.txt"
+    if text_path.exists():
+        with open(text_path, 'r', encoding='utf-8') as f:
+            anon_text = f.read()
+        
+        print(f"\n📝 TEXTE ANONYMISÉ (extrait):")
+        print("   " + "-"*76)
+        lines = anon_text.split('\n')[:8]
+        for line in lines:
+            if line.strip():
+                print(f"   {line[:76]}")
+        print("   " + "-"*76)
+    
+    # Scan de fuite
+    if redacted_pdf.exists() and audit_path.exists():
+        print(f"\n🔒 VALIDATION DE SÉCURITÉ")
+        
+        scanner = LeakScanner()
+        leak_report = scanner.scan(redacted_pdf, audit_path)
+        
+        if leak_report.is_safe:
+            print(f"   ✅ DOCUMENT SÛR")
+            print(f"      Aucune fuite de PII détectée")
+            print(f"      Le document peut être diffusé en toute sécurité")
+        else:
+            print(f"   ⚠️  ATTENTION - {leak_report.leak_count} fuite(s)")
+            for severity, count in leak_report.severity_counts.items():
+                print(f"      {severity}: {count}")
+    
+    # Résumé
+    print(f"\n" + "="*80)
+    print("✨ RÉSUMÉ")
+    print("="*80)
+    
+    print(f"\n✓ Document anonymisé avec succès")
+    print(f"✓ {len(page0_pii)} PII détectés et masqués")
+    print(f"✓ Propagation globale des noms sur tout le document")
+    print(f"✓ Validation de sécurité : AUCUNE FUITE")
+    
+    print(f"\n📂 Fichiers générés:")
+    print(f"   • PDF anonymisé (raster): {redacted_pdf.name}")
+    print(f"   • PDF anonymisé (vector): {base_name}.redacted_vector.pdf")
+    print(f"   • Texte anonymisé: {base_name}.pseudonymise.txt")
+    print(f"   • Audit détaillé: {base_name}.audit.jsonl")
+    
+    print(f"\n💡 Répertoire: {output_dir}")
+    
+    print(f"\n🎯 PROCHAINES ÉTAPES:")
+    print(f"   1. Annoter manuellement ce document")
+    print(f"   2. Comparer avec l'évaluateur de qualité")
+    print(f"   3. Calculer Précision, Rappel, F1-Score")
+    print(f"   4. Identifier les améliorations possibles")
+    
+    print(f"\n" + "="*80)
+
+if __name__ == "__main__":
+    show_comparison()

test_anonymization_example.py

@@ -0,0 +1,130 @@
+#!/usr/bin/env python3
+"""
+Test d'anonymisation sur un document réel avec analyse des résultats.
+"""
+import sys
+from pathlib import Path
+
+# Importer le système d'anonymisation
+from anonymizer_core_refactored_onnx import anonymize_pdf_file
+
+# Importer les outils d'évaluation
+from evaluation import LeakScanner
+
+def main():
+    # Sélectionner un document simple
+    pdf_path = Path("tests/ground_truth/pdfs/003_simple_compte_rendu_CRO_23155084.pdf")
+    
+    if not pdf_path.exists():
+        print(f"✗ Document introuvable: {pdf_path}")
+        return 1
+    
+    print("="*80)
+    print("TEST D'ANONYMISATION SUR UN DOCUMENT RÉEL")
+    print("="*80)
+    print(f"\n📄 Document: {pdf_path.name}")
+    print(f"   Type: Compte-rendu opératoire (CRO)")
+    print(f"   Complexité: Simple (1 page)")
+    
+    # Créer le répertoire de sortie
+    output_dir = Path("tests/ground_truth/pdfs/anonymized_test")
+    output_dir.mkdir(exist_ok=True)
+    
+    print(f"\n📁 Répertoire de sortie: {output_dir}")
+    
+    # Anonymiser le document
+    print("\n🔄 Anonymisation en cours...")
+    print("   (Cela peut prendre quelques secondes)")
+    
+    try:
+        result = anonymize_pdf_file(
+            pdf_path=str(pdf_path),
+            output_dir=str(output_dir),
+            use_ner=True,  # Activer le NER
+            use_vlm=False,  # Désactiver le VLM pour ce test (plus rapide)
+            raster_dpi=150,
+            force_raster=False
+        )
+        
+        print("\n✓ Anonymisation terminée !")
+        
+        # Afficher les résultats
+        if result:
+            audit_path = output_dir / f"{pdf_path.stem}.audit.jsonl"
+            redacted_pdf = output_dir / f"{pdf_path.stem}.redacted_raster.pdf"
+            
+            if not redacted_pdf.exists():
+                redacted_pdf = output_dir / f"{pdf_path.stem}.redacted_vector.pdf"
+            
+            print(f"\n📊 Fichiers générés:")
+            print(f"   - PDF anonymisé: {redacted_pdf.name}")
+            print(f"   - Audit: {audit_path.name}")
+            
+            # Compter les PII détectés
+            if audit_path.exists():
+                with open(audit_path, 'r', encoding='utf-8') as f:
+                    pii_count = sum(1 for line in f if line.strip())
+                
+                print(f"\n🔍 PII détectés: {pii_count}")
+                
+                # Afficher les premiers PII
+                print("\n📋 Premiers PII détectés:")
+                with open(audit_path, 'r', encoding='utf-8') as f:
+                    for i, line in enumerate(f):
+                        if i >= 10:  # Limiter à 10
+                            break
+                        if line.strip():
+                            import json
+                            pii = json.loads(line)
+                            print(f"   {i+1}. {pii.get('kind', 'UNKNOWN'):15s} : {pii.get('original', '')[:50]}")
+                
+                if pii_count > 10:
+                    print(f"   ... et {pii_count - 10} autres")
+                
+                # Scanner les fuites
+                print("\n🔒 Scan de fuite en cours...")
+                scanner = LeakScanner()
+                
+                if redacted_pdf.exists():
+                    leak_report = scanner.scan(redacted_pdf, audit_path)
+                    
+                    if leak_report.is_safe:
+                        print("   ✓ DOCUMENT SÛR - Aucune fuite détectée")
+                    else:
+                        print(f"   ✗ ATTENTION - {leak_report.leak_count} fuite(s) détectée(s)")
+                        
+                        # Afficher les fuites par sévérité
+                        for severity, count in sorted(leak_report.severity_counts.items()):
+                            print(f"     - {severity}: {count}")
+                        
+                        # Afficher les premières fuites
+                        print("\n   Détails des fuites:")
+                        for i, leak in enumerate(leak_report.leaks[:5], 1):
+                            print(f"     {i}. [{leak['severity']}] {leak['message']}")
+                        
+                        if leak_report.leak_count > 5:
+                            print(f"     ... et {leak_report.leak_count - 5} autres")
+                else:
+                    print("   ⚠ PDF anonymisé introuvable, impossible de scanner")
+            
+            print("\n" + "="*80)
+            print("✨ Test terminé avec succès !")
+            print("="*80)
+            
+            print(f"\n📂 Fichiers disponibles dans: {output_dir}")
+            print("\n💡 Pour voir le PDF anonymisé:")
+            print(f"   xdg-open {redacted_pdf}")
+            
+            return 0
+        else:
+            print("\n✗ Erreur lors de l'anonymisation")
+            return 1
+            
+    except Exception as e:
+        print(f"\n✗ Erreur: {e}")
+        import traceback
+        traceback.print_exc()
+        return 1
+
+if __name__ == "__main__":
+    sys.exit(main())

tests/ground_truth/pdfs/anonymized_test/003_simple_compte_rendu_CRO_23155084.audit.jsonl

@@ -0,0 +1,25 @@
+{"page": 0, "kind": "NOM", "original": "GASTON GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": 0, "kind": "ADRESSE", "original": "10 RUE DES HAUTRS VENTS", "placeholder": "[ADRESSE]", "bbox_hint": null}
+{"page": 0, "kind": "CODE_POSTAL", "original": "14190 OUILLY LE TESSON", "placeholder": "[CODE_POSTAL]", "bbox_hint": null}
+{"page": 0, "kind": "DATE_NAISSANCE", "original": "Né le 02/04/2010", "placeholder": "[DATE_NAISSANCE]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GONTIER", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GONTIER", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "QUEANT", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_EXTRACTED", "original": "QUEANT", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_GLOBAL", "original": "GASTON", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_GLOBAL", "original": "GONTIER", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_GLOBAL", "original": "GILLES", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "NOM_GLOBAL", "original": "QUEANT", "placeholder": "[NOM]", "bbox_hint": null}
+{"page": -1, "kind": "ADRESSE_GLOBAL", "original": "10 RUE DES HAUTRS VENTS", "placeholder": "[ADRESSE]", "bbox_hint": null}
+{"page": -1, "kind": "CODE_POSTAL_GLOBAL", "original": "14190 OUILLY LE TESSON", "placeholder": "[CODE_POSTAL]", "bbox_hint": null}
+{"page": -1, "kind": "DATE_NAISSANCE_GLOBAL", "original": "Né le 02/04/2010", "placeholder": "[DATE_NAISSANCE]", "bbox_hint": null}