Stop an edge artefact and a speck filter from destroying a song

Olukainen juomukainen came out unusable, from two separate faults.

The scanner left a dark line down the sheet edge, running the full height
of every page but the first. Being taller than any bracket it won every
overlap in anchor selection and swallowed the page into one system, so
five pages of six proposed no cuts at all. A page's brackets and barlines
are all about one system tall, so a stroke far taller than the typical
one is not notation — relative to the page's own strokes, since a page
holding one big system is legitimate.

The trim then removed each system's bottom line of lyrics. It judged ink
blobs by area, and a letter is nowhere near the threshold; a whole line
of them is dozens of blobs, none of which qualifies. Measure ink per row
and per column instead — a line of text carries plenty in total, and a
fleck's row carries almost none, which is the case the filter was for.

Detection now finds three systems on every page of that score, and every
slice keeps all four voices' words.

noteman_slicer/render.py

@@ -26,7 +26,10 @@ from .project import Cut, Project
 
 MAX_WIDTH = 1920
 ALPHA_LEVELS = 16  # quantising alpha costs nothing visible and ~32% of the bytes
-_SPECK_AREA = 300  # ink blobs smaller than this don't anchor a trim
+# A row or column carrying less ink than this is a fleck, not content: at least
+# this many pixels, and at least this share of the slice's own size.
+_SPECK_INK = 8
+_SPECK_SHARE = 0.005
 
 
 @dataclass
@@ -105,26 +108,24 @@ def _ink_bbox(gray: np.ndarray) -> tuple[int, int, int, int] | None:
 
     One scan fleck at the far left would otherwise anchor the trim and shift
     that slice relative to every other one.
+
+    Measured per row and per column rather than per blob. Judging each blob on
+    its own area throws away a whole line of lyrics — every letter is its own
+    small component, and no single one is big enough to keep — which is how a
+    slice loses its bottom voice's words. A row carrying a line of text carries
+    plenty of ink *in total*, and a fleck's row carries almost none.
     """
-    ink = (gray < 200).astype(np.uint8)
-    count, _, stats, _ = cv2.connectedComponentsWithStats(ink, 8)
-    boxes = [
-        (
-            stats[i, cv2.CC_STAT_LEFT],
-            stats[i, cv2.CC_STAT_TOP],
-            stats[i, cv2.CC_STAT_LEFT] + stats[i, cv2.CC_STAT_WIDTH],
-            stats[i, cv2.CC_STAT_TOP] + stats[i, cv2.CC_STAT_HEIGHT],
-        )
-        for i in range(1, count)
-        if stats[i, cv2.CC_STAT_AREA] >= _SPECK_AREA
-    ]
-    if not boxes:
+    ink = gray < 200
+    rows, cols = ink.sum(axis=1), ink.sum(axis=0)
+    kept_rows = np.where(rows >= max(_SPECK_INK, ink.shape[1] * _SPECK_SHARE))[0]
+    kept_cols = np.where(cols >= max(_SPECK_INK, ink.shape[0] * _SPECK_SHARE))[0]
+    if not kept_rows.size or not kept_cols.size:
         return None
     return (
-        min(b[0] for b in boxes),
-        min(b[1] for b in boxes),
-        max(b[2] for b in boxes),
-        max(b[3] for b in boxes),
+        int(kept_cols[0]),
+        int(kept_rows[0]),
+        int(kept_cols[-1]) + 1,
+        int(kept_rows[-1]) + 1,
     )