From c91bffe5eab23f88b24179b5d8973b97835a5b4d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Antoni=20Nu=C3=B1ez=20Romeu?= <antoni.nunez@ficosa.com>
Date: Mon, 6 Jul 2026 15:10:58 +0200
Subject: [PATCH] fix: improve OCR CIP extraction to filter common TSI card
 text

The extractCip function was incorrectly matching 'TARJETASANITARIA'
instead of the actual barcode/CIP number because it's 16 alphanumeric
characters that matched the regex.

Added filtering for common Spanish text on TSI cards and prioritized
numeric-heavy patterns typical of barcode codes.
---
 backend/server.js | 67 +++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 62 insertions(+), 5 deletions(-)

diff --git a/backend/server.js b/backend/server.js
index 1a45bcc..8b9846b 100644
--- a/backend/server.js
+++ b/backend/server.js
@@ -531,12 +531,63 @@ const upload = multer({
 // Extract CIP from OCR text — looks for a long alphanumeric token (16 chars typical for CIP)
 function extractCip(text) {
   const cleaned = text.replace(/\s+/g, '');
-  // Try exact 16-char alphanumeric match first
+
+  // Common Spanish text on TSI cards that should NOT be treated as CIP
+  const CARD_TEXT_PATTERNS = [
+    /TARJETASANITARIA/i,
+    /TARJETASANIT/i,
+    /TARJETASEGURIDAD/i,
+    /SISTEMASALUD/i,
+    /SERVICIOANDALUZ/i,
+    /SALUD/i,
+    /NUMERO/i,
+    /NUM/i,
+    /CIP/i,
+    /CODIGO/i,
+    /IDENTIFICACION/i,
+    /PERSONAL/i,
+    /TITULAR/i,
+    /FECHA/i,
+    /NACIMIENTO/i,
+    /CADUCIDAD/i,
+    /VALIDA/i,
+    /ESPANA/i,
+    /ESPAÑA/i,
+  ];
+
+  // Check if OCR result is just common card text
+  const isCardText = (candidate) => {
+    const upper = candidate.toUpperCase();
+    return CARD_TEXT_PATTERNS.some(pattern => pattern.test(upper));
+  };
+
+  // Try to find barcodes first (typically numeric-heavy)
+  // TSI barcodes are usually numeric or alphanumeric with mostly numbers
+  const numericMatches = cleaned.match(/\d{10,30}/g) || [];
+  for (const match of numericMatches) {
+    if (match.length >= 10 && match.length <= 30) {
+      return match.toUpperCase();
+    }
+  }
+
+  // Try exact 16-char alphanumeric match, but filter card text
   const exact = cleaned.match(/[A-Z0-9]{16}/i);
-  if (exact) return exact[0].toUpperCase();
-  // Fallback: 8–30 alphanumeric chars
-  const loose = cleaned.match(/[A-Z0-9]{8,30}/i);
-  if (loose) return loose[0].toUpperCase();
+  if (exact) {
+    const candidate = exact[0].toUpperCase();
+    if (!isCardText(candidate)) {
+      return candidate;
+    }
+  }
+
+  // Fallback: 8–30 alphanumeric chars, filter card text
+  const looseMatches = cleaned.match(/[A-Z0-9]{8,30}/gi) || [];
+  for (const match of looseMatches) {
+    const candidate = match.toUpperCase();
+    if (!isCardText(candidate)) {
+      return candidate;
+    }
+  }
+
   return null;
 }
 
@@ -546,11 +597,17 @@ app.post('/api/tsi/ocr', upload.single('photo'), async (req, res) => {
       return res.status(400).json({ error: 'No image file provided' });
     }
 
+    console.log('[OCR] Processing image, size:', req.file.size, 'bytes');
+
     const worker = await createWorker('spa+eng');
     const { data } = await worker.recognize(req.file.buffer);
     await worker.terminate();
 
     console.log('[OCR] Raw text:', data.text);
+    console.log('[OCR] Words:', data.words?.length || 0, 'words detected');
+    if (data.words && data.words.length > 0) {
+      console.log('[OCR] First 10 words:', data.words.slice(0, 10).map(w => `${w.text}(${w.confidence?.toFixed(2)})`).join(', '));
+    }
     const cip = extractCip(data.text);
     console.log('[OCR] Extracted CIP:', cip);
 
-- 
2.52.0