ignore spurious text in clipping detection

microqa/engine.py

@@ -1,3 +1,4 @@
+import re
 from sys import stdout
 from typing import Optional
 
@@ -243,10 +244,7 @@ def compute_text_margin_px(
         Integer pixel count if text is present; otherwise `None`.
     """
 
-    word_margins_all_directions = (
-        np.sort(
-            np.concat(
-                [
+    block_margins = [
         np.array(
             [
                 block.x0,
@@ -256,10 +254,12 @@ def compute_text_margin_px(
             ]
         )
         for block in text_blocks
+        # Exclude text without clear alphanumeric substance.
+        if len(re.sub(r"[^a-zA-Z0-9]", "", block.text)) > 1
     ]
-            ).astype(np.int_)
-        )
-        if len(text_blocks) > 0
+    word_margins_all_directions = (
+        np.sort(np.concat(block_margins).astype(np.int_))
+        if len(block_margins) > 0
         else np.array([])
     )
     # Skip the n closest words to the edge, to help ignore stray OCR artifacts.