Update app.py

app.py

@@ -60,12 +60,24 @@ def test_fft_frequency(roi):
     h, w = mag.shape
     inner = mag[h // 3:2 * h // 3, w // 3:2 * w // 3].mean()
     outer = mag.mean()
-    return 0.90 if (inner / outer) < 1.5 else 0.40
+    ratio = inner / outer
+    # v5.1 : seuil durci (1.3 au lieu de 1.5) — deepfakes ont spectre trop lisse
+    return 0.90 if ratio < 1.3 else 0.40
 
 # ═══════════════════════════════════════════════════════════
-# ÉTAPE 3 : NOUVEAUX MOTEURS V5 (Sora 2.0 / Runway / Kling)
+# ÉTAPE 3 : MOTEURS V5.1 — recalibrés face aux deepfakes WhatsApp/Sora
 # ═══════════════════════════════════════════════════════════
 def test_optical_flow(frames):
+    """
+    v5.1 — Recalibration clé :
+    Les deepfakes modernes (Sora 2.0, Wan, Kling) ont un flux optique
+    avec ratio std/mean proche de 1.0 → trop uniforme par rapport au
+    mouvement humain naturel (ratio attendu : 1.5–3.5).
+    Seuils resserrés :
+      - ratio < 0.80 ou 0.85 < ratio < 1.20  → suspect (trop lisse/uniforme)
+      - ratio 1.20–1.50 → ambigu
+      - ratio > 1.50 → naturel
+    """
     if len(frames) < 2:
         return 0.50
     scores = []
@@ -79,17 +91,21 @@ def test_optical_flow(frames):
         mag_std = np.std(magnitude)
         mag_mean = np.mean(magnitude)
         if mag_mean < 0.01:
-            scores.append(0.60)
+            scores.append(0.55)
         else:
             ratio = mag_std / mag_mean
-            if 0.30 < ratio < 2.50:
-                scores.append(0.88)   # Mouvement naturel
+            if ratio > 1.60:
+                scores.append(0.88)   # Mouvement naturel irrégulier
+            elif ratio > 1.30:
+                scores.append(0.62)   # Ambigu
+            elif ratio > 0.80:
+                scores.append(0.18)   # ★ Zone deepfake typique (ratio ~1.0) ★
             elif ratio < 0.15:
-                scores.append(0.18)   # Trop lisse → IA
+                scores.append(0.15)   # Trop lisse → IA
             elif ratio > 4.0:
                 scores.append(0.22)   # Saccades → Injection
             else:
-                scores.append(0.50)
+                scores.append(0.35)
     return float(np.mean(scores)) if scores else 0.50
 
 def test_eye_region(frame):
@@ -147,15 +163,21 @@ def test_texture_lbp(roi):
         return 0.22
 
 # ═══════════════════════════════════════════════════════════
-# ÉTAPE 4 : VERDICT CALIBRÉ V5
+# ÉTAPE 4 : VERDICT v5.1 — seuils relevés
 # ═══════════════════════════════════════════════════════════
 def get_verdict(score_pct):
-    if score_pct >= 72:
+    """
+    v5.1 : seuils durcis.
+    < 55%  → DEEPFAKE DÉTECTÉ
+    55–72% → SUSPECT
+    > 72%  → AUTHENTIQUE
+    """
+    if score_pct > 72:
         return "✅ AUTHENTIQUE", "Cohérence temporelle, colorimétrique et texturale conforme."
-    elif score_pct >= 58:
+    elif score_pct >= 55:
         return "⚠️ SUSPECT", "Incohérences détectées — vérification manuelle recommandée."
     else:
-        return "🚨 DEEPFAKE DÉTECTÉ", "Anomalie majeure (IA générative : Sora/Runway/Kling détectés)."
+        return "🚨 DEEPFAKE DÉTECTÉ", "Anomalie majeure détectée (IA générative : Sora/Wan/Kling/Runway)."
 
 # ═══════════════════════════════════════════════════════════
 # ÉTAPE 5 : PIPELINE PRINCIPAL
@@ -202,14 +224,15 @@ def analyze_video(video_path):
             s6 = test_color_coherence(roi, frame)
             s7 = test_texture_lbp(roi)
             
+            # v5.1 : poids optical flow augmenté 25→35%, boundary réduit 10→5%
             final = (
-                s1 * 0.10 +   
-                s2 * 0.18 +   
-                s3 * 0.12 +   
-                flow_score * 0.25 +  
-                s5 * 0.10 +   
-                s6 * 0.15 +   
-                s7 * 0.10     
+                s1 * 0.05 +          # boundary
+                s2 * 0.15 +          # noise coherence
+                s3 * 0.10 +          # FFT
+                flow_score * 0.35 +  # ★ optical flow (moteur principal)
+                s5 * 0.10 +          # eye region
+                s6 * 0.15 +          # color LAB
+                s7 * 0.10            # LBP texture
             )
             per_face_scores.append(final)
             engine_log.append({
@@ -234,7 +257,7 @@ def analyze_video(video_path):
     verdict, explication = get_verdict(global_score_pct)
     sep = "─" * 52
     rapport = (
-        f"🛡️ VideoShield v5.0 — Rapport d'Authenticité\n{sep}\n"
+        f"🛡️ VideoShield v5.1 — Rapport d'Authenticité\n{sep}\n"
         f"VERDICT   : {verdict}\n"
         f"SCORE     : {global_score_pct}%\n"
         f"ANALYSE   : {explication}\n"
@@ -242,13 +265,14 @@ def analyze_video(video_path):
         f"{sep}\n"
         f"Moteurs   : Boundary | Noise | FFT | Optical Flow\n"
         f"          : Eye Region | Color LAB | LBP Texture\n"
-        f"Cibles    : Sora 2.0 | Runway Gen-3 | Kling | Pika 2\n"
+        f"Cibles    : Sora 2.0 | Wan 2.1 | Kling | Pika 2 | Runway Gen-3\n"
         f"Standard  : IASA TC-04 | ACoNum Tunisia 2026\n"
+        f"Version   : v5.1 — recalibrée juin 2026\n"
         f"{sep}"
     )
     
     res_json = {
-        "version": "VideoShield v5.0",
+        "version": "VideoShield v5.1",
         "score": global_score_pct,
         "verdict": verdict,
         "flow_global": round(flow_score, 3),
@@ -287,13 +311,14 @@ def generate_iscc(video_path):
         return f"❌ Erreur ISCC : {str(e)}"
 
 # ═══════════════════════════════════════════════════════════
-# INTERFACE GRADIO V5
+# INTERFACE GRADIO V5.1
 # ═══════════════════════════════════════════════════════════
-with gr.Blocks(title="VideoShield v5.0", theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🛡️ VideoShield v5.0 — Authenticité Vidéo IA")
+with gr.Blocks(title="VideoShield v5.1", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🛡️ VideoShield v5.1 — Authenticité Vidéo IA")
     gr.Markdown(
-        "Détection de deepfakes génératifs (Sora 2.0, Runway Gen-3, Kling, Pika 2) "
-        "via 7 moteurs forensiques OpenCV. Standard IASA TC-04 · ACoNum Tunisia 2026."
+        "Détection de deepfakes génératifs (Sora 2.0, Wan 2.1, Runway Gen-3, Kling, Pika 2) "
+        "via 7 moteurs forensiques OpenCV. Standard IASA TC-04 · ACoNum Tunisia 2026. "
+        "**v5.1** : flux optique recalibré — détection ratio uniforme ~1.0 (signature deepfake)."
     )
     
     with gr.Tab("🔍 Analyse Deepfake"):