feat: Add multiple TVA entries support for Romanian receipts

- Add TvaEntry schema supporting multiple TVA rates (A, B, C, D codes)
- Update OCR extractor to extract multiple TVA entries from receipts
- Support both old (19%, 9%, 5%) and new Romanian rates (21%, 11% from Aug 2025)
- Add tva_breakdown, tva_total, items_count, vendor_address to Receipt model
- Update OCRPreview.vue to display TVA entries with rate badges
- Add "Detalii Suplimentare" section in ReceiptCreateView with editable TVA table
- Add TVA breakdown display in ReceiptDetailView
- Create database migration for new TVA columns

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

data-entry-app/backend/app/services/image_preprocessor.py

@@ -23,24 +23,37 @@ class ImagePreprocessor:
             raise ValueError(f"Could not load image: {path}")
         return image
 
-    def pdf_to_images(self, path: Path, dpi: int = 300) -> List[np.ndarray]:
-        """Convert PDF to images."""
+    def pdf_to_images(self, path: Path, dpi: int = 400) -> List[np.ndarray]:
+        """
+        Convert PDF to images with high DPI for better OCR.
+
+        Args:
+            path: Path to PDF file
+            dpi: Resolution (400 recommended for receipts, higher = better quality but slower)
+        """
         if not PDF_AVAILABLE:
             raise RuntimeError("pdf2image not available. Install with: pip install pdf2image")
+        # Use 400 DPI for better text recognition on thermal receipts
         images = pdf2image.convert_from_path(str(path), dpi=dpi)
         return [np.array(img) for img in images]
 
-    def preprocess(self, image: np.ndarray) -> np.ndarray:
+    def preprocess(self, image: np.ndarray, high_quality: bool = True) -> np.ndarray:
         """
         Apply preprocessing pipeline for thermal receipt images.
 
         Pipeline:
         1. Convert to grayscale
-        2. Resize if too small (min 1000px width)
+        2. Resize if too small (min 1500px width for high quality)
         3. Deskew (straighten rotated text)
-        4. Denoise (Non-local means)
-        5. Adaptive thresholding (binarization)
-        6. Morphological close (connect broken chars)
+        4. Contrast enhancement (CLAHE)
+        5. Denoise (Non-local means)
+        6. Sharpening (for clearer text edges)
+        7. Adaptive thresholding (binarization)
+        8. Morphological operations (connect broken chars)
+
+        Args:
+            image: Input image (BGR or grayscale)
+            high_quality: If True, apply more aggressive preprocessing
         """
         # 1. Grayscale
         if len(image.shape) == 3:
@@ -48,10 +61,11 @@ class ImagePreprocessor:
         else:
             gray = image.copy()
 
-        # 2. Resize if too small
+        # 2. Resize if too small (larger = better OCR)
         height, width = gray.shape
-        if width < 1000:
-            scale = 1000 / width
+        min_width = 1500 if high_quality else 1000
+        if width < min_width:
+            scale = min_width / width
             gray = cv2.resize(
                 gray, None, fx=scale, fy=scale,
                 interpolation=cv2.INTER_CUBIC
@@ -60,24 +74,43 @@ class ImagePreprocessor:
         # 3. Deskew
         gray = self._deskew(gray)
 
-        # 4. Denoise
+        # 4. Contrast enhancement with CLAHE (Contrast Limited Adaptive Histogram Equalization)
+        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+        enhanced = clahe.apply(gray)
+
+        # 5. Denoise (slightly less aggressive to preserve text details)
         denoised = cv2.fastNlMeansDenoising(
-            gray, h=10,
+            enhanced, h=8,  # Lower h = preserve more details
             templateWindowSize=7,
             searchWindowSize=21
         )
 
-        # 5. Adaptive thresholding
+        # 6. Sharpening to enhance text edges
+        if high_quality:
+            # Unsharp mask for better text clarity
+            gaussian = cv2.GaussianBlur(denoised, (0, 0), 2.0)
+            sharpened = cv2.addWeighted(denoised, 1.5, gaussian, -0.5, 0)
+        else:
+            sharpened = denoised
+
+        # 7. Adaptive thresholding with optimized parameters
         binary = cv2.adaptiveThreshold(
-            denoised, 255,
+            sharpened, 255,
             cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
             cv2.THRESH_BINARY,
-            blockSize=15, C=8
+            blockSize=11,  # Smaller block = better for small text
+            C=5  # Lower C = darker result, better for faded receipts
         )
 
-        # 6. Morphological close
-        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
-        result = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)
+        # 8. Morphological operations
+        # Close small gaps in characters
+        kernel_close = cv2.getStructuringElement(cv2.MORPH_RECT, (2, 2))
+        result = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel_close)
+
+        # Optional: Remove small noise spots
+        if high_quality:
+            kernel_open = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 1))
+            result = cv2.morphologyEx(result, cv2.MORPH_OPEN, kernel_open)
 
         return result