Fix macro false positive on landscapes: sharpness_ratio > 3.0 is now a hard gating requirement for MACRO_DSLR

agents/modality_detector.py

@@ -1,12 +1,12 @@
 """
-FORENSIQ — Capture Modality Detector v3
+FORENSIQ — Capture Modality Detector v4
 
 Classifies images BEFORE forensic analysis. Pure content-based detection
 that works even when Gradio strips metadata (format=None, no EXIF).
 
-Key fix v3: Relaxed macro detection thresholds. blur_frac uses adaptive 
-threshold (p40 instead of p25) for more accurate blur region estimation.
-Added blur_uniformity as 5th macro indicator. Debug logging to stderr.
+v4 fix: sharpness_ratio > 3.0 is now a HARD requirement for macro detection.
+Without extreme center-vs-edge sharpness, other indicators (bimodal ratio,
+background uniformity) are not sufficient to distinguish macro from landscape.
 """
 
 import sys
@@ -70,14 +70,13 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     sharp_region = sharpness > sharp_thresh
     sharp_frac = float(np.mean(sharp_region))
     
-    # Blur region: use ADAPTIVE threshold (p40 of sharpness)
-    # This gives a more meaningful blur fraction than the fixed p25 definition
-    # which is always ~25% by construction.
-    blur_adaptive_thresh = float(np.percentile(sharpness, 40))
-    blur_region = sharpness < blur_adaptive_thresh
+    # Blur region: use content-based threshold (20% of p95)
+    # This is more meaningful than percentile-based — it measures actual blur
+    blur_content_thresh = p95 * 0.20
+    blur_region = sharpness < blur_content_thresh
     blur_frac = float(np.mean(blur_region))
     
-    # Blur uniformity (computational blur is very uniform)
+    # Blur uniformity (computational/optical blur is very uniform)
     blur_vals = sharpness[blur_region] if np.any(blur_region) else np.array([1])
     blur_uniformity = 1.0 - min(float(np.std(blur_vals)) / (float(np.mean(blur_vals)) + 1e-9), 1.0)
     
@@ -98,7 +97,6 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     indicators["has_detail"] = has_detail
     
     # ── Portrait mode detection ───────────────────────────────────────
-    # CRITICAL: If no Bayer pattern, this CANNOT be a smartphone portrait photo.
     can_be_portrait = has_detail and has_bayer
     
     portrait_score = 0.0
@@ -116,10 +114,11 @@ def detect_modality(img: Image.Image) -> ModalityResult:
         indicators["portrait_detected"] = True
     
     # ── Macro/DSLR shallow DoF detection ─────────────────────────────
-    # Macro photos: extreme sharpness ratio (center vs edge), very high
-    # bimodal ratio, large blur region, uniform bokeh, and optionally Bayer.
+    # KEY INSIGHT: The defining feature of macro photography is EXTREME
+    # center-vs-edge sharpness difference (>3x). Without this, an image
+    # cannot be macro — it could be a landscape with fog/haze that mimics
+    # bimodal sharpness distribution.
     
-    # Compute center-vs-edge sharpness ratio
     ch_s, cw_s = gray.shape
     center_region = sharpness[ch_s//4:3*ch_s//4, cw_s//4:3*cw_s//4]
     edge_region = np.concatenate([
@@ -132,7 +131,7 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     edge_sharp = float(np.mean(edge_region))
     sharpness_ratio = center_sharp / (edge_sharp + 1e-9)
     
-    # Background uniformity: how uniform is the blurred region's color?
+    # Background uniformity
     blur_region_pixels = gray[blur_region] if np.any(blur_region) else np.array([128])
     bg_color_std = float(np.std(blur_region_pixels))
     
@@ -141,36 +140,42 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     indicators["center_sharp_p90"] = round(center_sharp, 2)
     indicators["edge_sharp_mean"] = round(edge_sharp, 2)
     
-    # ── Macro scoring (5 indicators) ─────────────────────────────────
+    # ── Macro scoring (GATED by sharpness_ratio) ─────────────────────
+    # sharpness_ratio > 3.0 is a HARD REQUIREMENT — other signals alone
+    # are not sufficient to distinguish macro from landscape/portrait.
+    
     macro_score = 0.0
     macro_components = []
     
-    if has_detail and sharpness_ratio > 3.0:
+    macro_gate_passed = has_detail and sharpness_ratio > 3.0
+    
+    if macro_gate_passed:
+        # Base score for passing the hard gate
         macro_score += 0.25
         macro_components.append(f"ratio={sharpness_ratio:.1f}")
-    
-    if has_detail and blur_frac > 0.30:
-        # Relaxed from 0.35 to 0.30 — adaptive p40 threshold gives ~40% for macro
-        macro_score += 0.15
-        macro_components.append(f"blur={blur_frac:.2f}")
-    
-    if has_detail and bimodal_ratio > 1.5:
-        macro_score += 0.20
-        macro_components.append(f"bimodal={bimodal_ratio:.2f}")
-    
-    if has_detail and bg_color_std < 40:
-        macro_score += 0.15
-        macro_components.append(f"bg_std={bg_color_std:.1f}")
-    
-    # NEW: blur uniformity — bokeh backgrounds are very uniform in blur intensity
-    if has_detail and blur_uniformity > 0.6:
-        macro_score += 0.15
-        macro_components.append(f"blur_uni={blur_uniformity:.2f}")
+        
+        # Additional indicators (each adds confidence)
+        if blur_frac > 0.25:
+            macro_score += 0.15
+            macro_components.append(f"blur={blur_frac:.2f}")
+        
+        if bimodal_ratio > 1.5:
+            macro_score += 0.20
+            macro_components.append(f"bimodal={bimodal_ratio:.2f}")
+        
+        if bg_color_std < 40:
+            macro_score += 0.15
+            macro_components.append(f"bg_std={bg_color_std:.1f}")
+        
+        if blur_uniformity > 0.6:
+            macro_score += 0.15
+            macro_components.append(f"blur_uni={blur_uniformity:.2f}")
     
     indicators["macro_score"] = round(macro_score, 3)
     indicators["macro_components"] = macro_components
+    indicators["macro_gate_passed"] = macro_gate_passed
     
-    # Macro requires Bayer OR high signal strength (Unsplash strips Bayer traces)
+    # Macro requires GATE + (Bayer OR high signal strength)
     if macro_score >= 0.55:
         scores["MACRO_DSLR"] = macro_score
         indicators["macro_detected"] = True
@@ -274,13 +279,11 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     
     # SAFETY GUARD 2: No Bayer pattern = not from a real camera sensor.
     # Exception: MACRO_DSLR with strong signals can bypass this
-    # (Unsplash CDN processing can strip Bayer traces from real DSLR photos)
     if not has_bayer and modality in ("PORTRAIT_MODE", "SMARTPHONE", "MESSAGING"):
         modality = "UNKNOWN"
         conf = 0.2
         indicators["safety_override"] = f"No Bayer CFA pattern (margin={bayer_margin:.3f}) — not from a real camera sensor. All suppression disabled."
     elif not has_bayer and modality == "MACRO_DSLR" and scores.get("MACRO_DSLR", 0) < 0.55:
-        # Weak macro signal without Bayer — don't trust it
         modality = "UNKNOWN"
         conf = 0.2
         indicators["safety_override"] = f"Macro signal weak ({scores.get('MACRO_DSLR', 0):.2f}) + no Bayer — suppression disabled"
@@ -288,7 +291,7 @@ def detect_modality(img: Image.Image) -> ModalityResult:
     # ═══ DEBUG LOGGING ════════════════════════════════════════════════
     print(f"[MODALITY] detected={modality} conf={conf:.2f} scores={scores}", file=sys.stderr)
     print(f"[MODALITY] has_bayer={has_bayer} bayer_margin={bayer_margin:.4f}", file=sys.stderr)
-    print(f"[MODALITY] macro_score={macro_score:.3f} components={macro_components}", file=sys.stderr)
+    print(f"[MODALITY] macro_score={macro_score:.3f} gate={macro_gate_passed} components={macro_components}", file=sys.stderr)
     print(f"[MODALITY] sharpness_ratio={sharpness_ratio:.2f} bimodal={bimodal_ratio:.3f} blur_frac={blur_frac:.3f} blur_uni={blur_uniformity:.3f} bg_std={bg_color_std:.2f}", file=sys.stderr)
     print(f"[MODALITY] p95={p95:.2f} has_detail={has_detail}", file=sys.stderr)
     if indicators.get("safety_override"):
@@ -320,22 +323,18 @@ def detect_modality(img: Image.Image) -> ModalityResult:
 def _get_modality_adjustments(modality: str) -> dict:
     if modality == "MACRO_DSLR":
         return {
-            # Extreme shallow DoF creates uniform bokeh → high autocorrelation
             "Autocorrelation Peak": 0.1,
             "Texture Repetition": 0.1,
             "DoF Consistency": 0.1,
-            # JPEG delivery pipeline removes sensor traces
             "Bayer CFA Pattern": 0.3,
             "CFA Nyquist": 0.3,
             "PRNU Uniformity": 0.2,
             "Demosaic Interpolation": 0.4,
-            # Bimodal content (sharp subject + smooth bokeh) creates extreme kurtosis
             "DCT Kurtosis": 0.1,
             "Wavelet Kurtosis": 0.1,
             "Spectral Slope 1/f²": 0.4,
             "Spectral Symmetry": 0.3,
             "Phase Coherence": 0.4,
-            # Noise inconsistency between sharp subject and bokeh
             "Noise Spatial Frequency": 0.2,
             "Poisson-Gaussian Model": 0.2,
             "HF Noise Structure": 0.3,