Barcode detection causes false positives with plain numeric text

[shukebeta]

Description

The barcode/QR code detection feature (enabled by default with detect-codes=true) can produce false positives when processing images containing plain numeric text. When a false positive occurs, the OCR path is completely skipped, resulting in incorrect text extraction.

Steps to Reproduce

1. Ensure detect-codes=true in settings (default)
2. Capture a screenshot containing numeric text that resembles a barcode pattern (e.g., "91385057399027")
3. NormCap detects it as a barcode and returns only the numbers
4. Expected OCR processing with proper text extraction is skipped

Example Log

WARNING - === CODES detection enabled, calling detect_codes() ===
WARNING - === Found 1 raw results, codes=['91385057399027'] ===
WARNING - === Single code detected: '91385057399027' (type=TextType.SINGLE_LINE) ===
WARNING - === detect_codes() returned: DetectionResult(...detector=<TextDetector.BARCODE>) ===

Current Behavior

When detect_codes=true (default), the detection logic prioritizes barcode/QR code detection:

• If a code is detected (even falsely), OCR is skipped entirely
• Users get incomplete/incorrect text extraction without realizing why

Expected Behavior

Potential improvements:

1. Reduce false positive rate: Configure zxingcpp with stricter thresholds
2. Confidence scoring: Only skip OCR if barcode confidence is high
3. Dual detection: Run both code detection and OCR, intelligently choose the better result
4. User control: Make it easier to disable code detection when not needed

Environment

• NormCap version: latest (main branch)
• Platform: Linux
• Settings: default (detect-codes=true)

Impact

This affects users processing:

• Mixed text with numbers (especially Asian text with digits)
• Documents with number sequences
• Any content where numeric patterns might trigger false barcode detection

Workaround

Manually disable code detection:

# In settings GUI: uncheck "Detect codes"
# Or edit config:
sed -i 's/detect-codes=true/detect-codes=false/' ~/.config/normcap/settings.conf

Related

This issue was discovered while testing PR #801 (smart whitespace stripping for CJK text).