Benjamin_Boenisch/breakpilot-lehrer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: merge overlapping OCR words with different text (Stick/Stück)

Browse Source 
Two issues in paddle-kombi word merge:

1. Overlap threshold too strict: PaddleOCR "Stick" and Tesseract
   "Stück" overlap at 48.6%, just below the 50% threshold. Both words
   ended up in the result, overlapping on the same position.
   Fix: lower threshold from 50% to 40%.

2. Text selection blind to confidence: always took PaddleOCR text
   even when Tesseract had higher confidence and correct text.
   Fix: when texts differ due to spatial-only match, prefer the
   engine with higher confidence.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-03-18 07:00:57 +01:00
parent fd79d5e4fa
commit a25214126d
1 changed files with 13 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
klausur-service/backend/ocr_pipeline_api.py
View File

@@ -3219,8 +3219,10 @@ def _merge_row_sequences(paddle_row: list, tess_row: list) -> list:
# Same text or one contains the other
is_same = (pt == tt) or (len(pt) > 1 and len(tt) > 1 and (pt in tt or tt in pt))
# Spatial overlap check: if words overlap >= 50% horizontally,
# they're the same physical word regardless of OCR text differences
# Spatial overlap check: if words overlap >= 40% horizontally,
# they're the same physical word regardless of OCR text differences.
# (40% catches borderline cases like "Stick"/"Stück" at 48% overlap)
spatial_match = False
if not is_same:
overlap_left = max(pw["left"], tw["left"])
overlap_right = min(
@@ -3229,8 +3231,9 @@ def _merge_row_sequences(paddle_row: list, tess_row: list) -> list:
)
overlap_w = max(0, overlap_right - overlap_left)
min_w = min(pw.get("width", 1), tw.get("width", 1))
if min_w > 0 and overlap_w / min_w >= 0.5:
if min_w > 0 and overlap_w / min_w >= 0.4:
is_same = True
spatial_match = True
if is_same:
# Matched — average coordinates weighted by confidence
@@ -3239,8 +3242,14 @@ def _merge_row_sequences(paddle_row: list, tess_row: list) -> list:
total = pc + tc
if total == 0:
total = 1
# Text: prefer higher-confidence engine when texts differ
# (e.g. Tesseract "Stück" conf=98 vs PaddleOCR "Stick" conf=80)
if spatial_match and pc < tc:
best_text = tw["text"]
else:
best_text = pw["text"]
merged.append({
"text": pw["text"], # Paddle text preferred
"text": best_text,
"left": round((pw["left"] * pc + tw["left"] * tc) / total),
"top": round((pw["top"] * pc + tw["top"] * tc) / total),
"width": round((pw["width"] * pc + tw["width"] * tc) / total),