fix: tighten illustration noise filter — alpha ratio, per-region cleanup

smoke_signal_tab.py

@@ -1721,13 +1721,41 @@ def _run_tesseract_batch(
             alpha_ratio = (letters / printable) if printable else 0.0
             tokens = [t for t in full_text.split() if t]
             avg_token_len = (sum(len(t) for t in tokens) / len(tokens)) if tokens else 0.0
-            if regions and avg_conf <= 0.42 and alpha_ratio < 0.62 and avg_token_len < 3.2:
+            # Filter illustration noise — two thresholds:
+            # 1. Strict: low conf + low alpha + short tokens (original heuristic, loosened)
+            if regions and avg_conf <= 0.55 and alpha_ratio < 0.72 and avg_token_len < 3.5:
                 return {
                     "regions": [],
                     "confidence": 0.0,
                     "method": "tesseract-noise-filtered",
                     "tesseract_lang": lang_hint or "eng",
                 }
+            # 2. Pure gibberish: very low alpha ratio regardless of confidence
+            if regions and alpha_ratio < 0.50:
+                return {
+                    "regions": [],
+                    "confidence": 0.0,
+                    "method": "tesseract-noise-filtered",
+                    "tesseract_lang": lang_hint or "eng",
+                }
+            # 3. Filter individual regions that look like illustration noise
+            # Keep only regions where alpha ratio >= 0.60 and token length >= 2
+            clean_regions = []
+            for r in regions:
+                txt = r["text"]
+                letters = sum(1 for c in txt if c.isalpha())
+                printable = sum(1 for c in txt if c.isprintable() and not c.isspace())
+                r_alpha = (letters / printable) if printable else 0.0
+                tokens = [t for t in txt.split() if t]
+                r_avg_tok = (sum(len(t) for t in tokens) / len(tokens)) if tokens else 0.0
+                if r_alpha >= 0.55 or r["confidence"] >= 0.75:
+                    clean_regions.append(r)
+            if clean_regions and len(clean_regions) < len(regions):
+                # Recalculate confidence without noise regions
+                cw = sum(r["confidence"] * r["word_count"] for r in clean_regions)
+                wc = sum(r["word_count"] for r in clean_regions)
+                avg_conf = round(cw / max(wc, 1), 4)
+                regions = clean_regions
 
             return {
                 "regions": regions,