From 84e1a6a82e6ea7893132d1f74debfe2947bbd2b6 Mon Sep 17 00:00:00 2001
From: Esa Kataja <ekataja@gmail.com>
Date: Tue, 3 Dec 2024 21:47:03 +0200
Subject: [PATCH] refactor: consolidate image trimming and standardize
 processing

- Replace OpenCV-based trim_whitespace with PIL implementation
- Remove redundant trim_whitespace_pil function
- Add initial trim step to extract command
- Standardize image processing to 2048px wide grayscale using Lanczos
---
 pdf_cleaner.py | 131 +++++++++++++++++++++++++------------------------
 1 file changed, 68 insertions(+), 63 deletions(-)

diff --git a/pdf_cleaner.py b/pdf_cleaner.py
index 2ca1494..d115c2d 100755
--- a/pdf_cleaner.py
+++ b/pdf_cleaner.py
@@ -14,6 +14,7 @@ from pdf2image import convert_from_path
 
 import settings
 from settings import OptimizationLevel
+from PIL import Image
 
 class Settings:
     TRIM_PADDING_PIXELS = 20
@@ -52,12 +53,79 @@ def extract(input_pdf):
     
     # Save each page
     for i, page in enumerate(pages):
+        # Convert to grayscale
+        page = page.convert('L')
+        
         output_path = os.path.join(temp_dir, f"page_{i:03d}.png")
+        
+        # Save initial version
+        page.save(output_path, "PNG", optimize=False)
+        
+        # Trim whitespace
+        trim_whitespace(output_path)
+        
+        # Reload the trimmed image
+        page = Image.open(output_path)
+        
+        # Calculate new height maintaining aspect ratio
+        width = 2048
+        ratio = width / page.width
+        height = int(page.height * ratio)
+        
+        # Resize using Lanczos
+        page = page.resize((width, height), Image.Resampling.LANCZOS)
+        
+        # Save final version
         page.save(output_path, "PNG", optimize=False)
         print(f"Saved page {i+1}/{len(pages)}")
     
     print(f"Extracted {len(pages)} pages to {temp_dir}/")
 
+def trim_whitespace(image_path: str) -> bool:
+    """Remove white space from around the image.
+    
+    Handles both RGB and RGBA images, treating transparent pixels as white.
+    
+    Args:
+        image_path: Path to the image file
+    
+    Returns:
+        bool: True if successful, False otherwise
+    """
+    try:
+        # Open image with PIL
+        image = Image.open(image_path)
+        
+        # Convert to grayscale if not already
+        if image.mode != 'L':
+            image = image.convert('L')
+        
+        # Get the bounding box of non-white pixels
+        bbox = image.getbbox()
+        if not bbox:
+            print(f"Warning: No content found in {image_path}")
+            return False
+        
+        # Add padding
+        padding = settings.TRIM_PADDING_PIXELS
+        width, height = image.size
+        x1, y1, x2, y2 = bbox
+        x1 = max(0, x1 - padding)
+        y1 = max(0, y1 - padding)
+        x2 = min(width, x2 + padding)
+        y2 = min(height, y2 + padding)
+        
+        # Crop the image
+        image = image.crop((x1, y1, x2, y2))
+        
+        # Save the image
+        image.save(image_path, "PNG", optimize=False)
+        return True
+        
+    except Exception as e:
+        print(f"Error processing {image_path}: {str(e)}")
+        return False
+
 @cli.command()
 def deskew():
     """Deskew all pages in temporary directory."""
@@ -132,69 +200,6 @@ def deskew():
         else:
             print("  No line segments detected")
 
-def trim_whitespace(image_path: str) -> bool:
-    """Remove white space from around the image.
-    
-    Handles both RGB and RGBA images, treating transparent pixels as white.
-    
-    Returns:
-        bool: True if successful, False otherwise
-    """
-    try:
-        # Read the image with alpha channel
-        img = cv2.imread(image_path, cv2.IMREAD_UNCHANGED)
-        if img is None:
-            print(f"Warning: Could not read image {image_path}")
-            return False
-        
-        # Convert to grayscale, handling transparency
-        if img.shape[-1] == 4:  # RGBA
-            # Create a white background
-            white_background = np.ones_like(img, dtype=np.uint8) * 255
-            
-            # Extract alpha channel and create mask
-            alpha = img[:, :, 3]
-            alpha_mask = alpha[:, :, np.newaxis] / 255.0
-            
-            # Blend image with white background based on alpha
-            img = (img[:, :, :3] * alpha_mask + white_background[:, :, :3] * (1 - alpha_mask)).astype(np.uint8)
-        
-        # Convert to grayscale
-        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        
-        # Threshold the image
-        _, thresh = cv2.threshold(gray, 250, 255, cv2.THRESH_BINARY_INV)
-        
-        # Find non-zero points
-        coords = cv2.findNonZero(thresh)
-        if coords is None:
-            print(f"Warning: No content found in {image_path}")
-            return False
-        
-        # Get bounding rectangle
-        x, y, w, h = cv2.boundingRect(coords)
-        
-        # Add padding
-        padding = settings.TRIM_PADDING_PIXELS
-        height, width = img.shape[:2]
-        x = max(0, x - padding)
-        y = max(0, y - padding)
-        w = min(width - x, w + 2 * padding)
-        h = min(height - y, h + 2 * padding)
-        
-        # Crop the image
-        if img.shape[-1] == 4:  # If original was RGBA
-            cropped = img[:, :, :4][y:y+h, x:x+w]  # Keep alpha channel
-        else:
-            cropped = img[y:y+h, x:x+w]
-        
-        # Save the cropped image with transparency preserved
-        cv2.imwrite(image_path, cropped)
-        return True
-    except Exception as e:
-        print(f"Error processing {image_path}: {str(e)}")
-        return False
-
 def convert_to_monochrome(image_path: str) -> bool:
     """Convert image to 1-bit monochrome.