fix(kreuzberg): add image preprocessing for better OCR quality
- Convert images to grayscale before OCR - Increase contrast by 2x for better text detection - This matches the preprocessing that effi-ocr (Tesseract) was using
This commit is contained in:
Folzo
@@ -0,0 +1,340 @@
|
||||
# Kreuzberg OCR Tool - Enhanced Version
|
||||
|
||||
## Installation
|
||||
|
||||
1. **Admin Panel** → **Tools** → **Create Tool**
|
||||
2. **Name**: `Kreuzberg OCR`
|
||||
3. **Description**: `Extract text from documents (PDF, images, scanned pages) using Kreuzberg OCR. Call `extract_file("filename")` on attached files.`
|
||||
4. **Coller le code ci-dessous**
|
||||
5. **Save and Activate**
|
||||
|
||||
## Tool Code
|
||||
|
||||
```python
|
||||
"""
|
||||
Kreuzberg OCR Tool for OpenWebUI
|
||||
Extracts text from scanned documents, images, and PDFs using Kreuzberg OCR.
|
||||
Supports: PDF, JPEG, PNG, TIFF, DOCX, PPTX, XLSX, EPUB, etc.
|
||||
"""
|
||||
|
||||
import os
|
||||
import glob
|
||||
import requests
|
||||
import hashlib
|
||||
import io
|
||||
from PIL import Image, ImageEnhance
|
||||
|
||||
|
||||
class Tools:
|
||||
def __init__(self):
|
||||
self.kreuzberg_url = os.getenv("KREUZBERG_URL", "http://192.168.1.11:8300")
|
||||
self.uploads_dir = "/app/backend/data/uploads"
|
||||
self.host_uploads_dir = "/home/effi/effi/data/uploads"
|
||||
|
||||
def _preprocess_image(self, image: Image.Image) -> Image.Image:
|
||||
"""Preprocess image for better OCR quality (grayscale + contrast)"""
|
||||
if image.mode in ('RGBA', 'LA', 'P'):
|
||||
image = image.convert('RGBA').convert('RGB')
|
||||
elif image.mode != 'RGB':
|
||||
image = image.convert('RGB')
|
||||
image = image.convert('L')
|
||||
enhancer = ImageEnhance.Contrast(image)
|
||||
image = enhancer.enhance(2.0)
|
||||
return image
|
||||
|
||||
def extract_file(self, filename: str) -> str:
|
||||
"""
|
||||
Extract text from a file using Kreuzberg OCR.
|
||||
|
||||
Args:
|
||||
filename: The filename shown in the chat (e.g., "document.pdf", "image.jpg")
|
||||
or the full name like "UUID_name.pdf"
|
||||
|
||||
Returns:
|
||||
Extracted text content with markdown formatting
|
||||
"""
|
||||
try:
|
||||
file_path = self._find_file(filename)
|
||||
|
||||
if not file_path:
|
||||
return f"❌ File not found: {filename}\n\nTip: Use `list_recent_files()` to see available files."
|
||||
|
||||
mime_type = self._get_mime_type(file_path)
|
||||
ext = os.path.splitext(file_path)[1].lower()
|
||||
is_image = ext in ('.jpg', '.jpeg', '.png', '.gif', '.bmp', '.tiff', '.tif', '.webp')
|
||||
|
||||
if is_image:
|
||||
with open(file_path, 'rb') as f:
|
||||
img = Image.open(f)
|
||||
img = self._preprocess_image(img)
|
||||
img_bytes = io.BytesIO()
|
||||
img.save(img_bytes, format='PNG')
|
||||
img_bytes.seek(0)
|
||||
files = {
|
||||
'files': (os.path.basename(file_path) + '.png', img_bytes, 'image/png')
|
||||
}
|
||||
else:
|
||||
with open(file_path, 'rb') as f:
|
||||
files = {
|
||||
'files': (os.path.basename(file_path), f, mime_type)
|
||||
}
|
||||
|
||||
data = {
|
||||
'image_export_mode': 'inline',
|
||||
'ocr': 'true'
|
||||
}
|
||||
|
||||
response = requests.post(
|
||||
f"{self.kreuzberg_url}/v1/convert/file",
|
||||
files=files,
|
||||
data=data,
|
||||
timeout=300
|
||||
|
||||
if response.status_code == 200:
|
||||
result = response.json()
|
||||
|
||||
if 'document' in result:
|
||||
content = result['document'].get('md_content', '')
|
||||
if content and len(content.strip()) > 10:
|
||||
# Clean up the content
|
||||
content = self._clean_content(content)
|
||||
|
||||
# Truncate if too long (LLM context limit)
|
||||
if len(content) > 15000:
|
||||
content = content[:15000] + "\n\n... (document truncated - too long for context)"
|
||||
|
||||
return f"📄 **{os.path.basename(file_path)}**\n\n{content}"
|
||||
else:
|
||||
return f"⚠️ No text extracted from {os.path.basename(file_path)}\n\nThe file may be empty, protected, or contain only images without selectable text."
|
||||
|
||||
return f"⚠️ Unexpected response format from Kreuzberg: {str(result)[:500]}"
|
||||
|
||||
elif response.status_code == 404:
|
||||
return f"❌ Kreuzberg endpoint not found. Is the server running at {self.kreuzberg_url}?"
|
||||
|
||||
else:
|
||||
return f"❌ Kreuzberg error ({response.status_code}): {response.text[:500]}"
|
||||
|
||||
except requests.Timeout:
|
||||
return f"⏱️ Timeout - File is too large or Kreuzberg is overloaded. Try a smaller file."
|
||||
|
||||
except requests.ConnectionError:
|
||||
return f"❌ Cannot connect to Kreuzberg at {self.kreuzberg_url}. Is the server running?"
|
||||
|
||||
except Exception as e:
|
||||
return f"❌ Error: {str(e)}"
|
||||
|
||||
def list_recent_files(self, count: int = 10) -> str:
|
||||
"""List recent uploaded files"""
|
||||
try:
|
||||
files = glob.glob(os.path.join(self.uploads_dir, "*"))
|
||||
|
||||
if not files and os.path.exists(self.host_uploads_dir):
|
||||
files = glob.glob(os.path.join(self.host_uploads_dir, "*"))
|
||||
|
||||
files = [f for f in files if os.path.isfile(f)]
|
||||
files = sorted(files, key=os.path.getmtime, reverse=True)[:count]
|
||||
|
||||
if not files:
|
||||
return "📂 No files found in uploads directory."
|
||||
|
||||
result = f"📂 **{len(files)} recent files:**\n\n"
|
||||
for f in files:
|
||||
name = os.path.basename(f)
|
||||
size = os.path.getsize(f)
|
||||
size_str = self._format_size(size)
|
||||
result += f"- `{name}` ({size_str})\n"
|
||||
|
||||
result += "\n💡 Use `extract_file(\"filename\")` to OCR a file."
|
||||
return result
|
||||
|
||||
except Exception as e:
|
||||
return f"❌ Error listing files: {str(e)}"
|
||||
|
||||
def search_files(self, pattern: str) -> str:
|
||||
"""Search for files matching a pattern"""
|
||||
try:
|
||||
# Search in uploads
|
||||
files = glob.glob(os.path.join(self.uploads_dir, f"*{pattern}*"))
|
||||
|
||||
if not files and os.path.exists(self.host_uploads_dir):
|
||||
files = glob.glob(os.path.join(self.host_uploads_dir, f"*{pattern}*"))
|
||||
|
||||
files = [f for f in files if os.path.isfile(f)]
|
||||
|
||||
if not files:
|
||||
return f"🔍 No files found matching: {pattern}"
|
||||
|
||||
result = f"🔍 **{len(files)} files found:**\n\n"
|
||||
for f in files[:20]: # Limit to 20
|
||||
name = os.path.basename(f)
|
||||
size = self._format_size(os.path.getsize(f))
|
||||
result += f"- `{name}` ({size})\n"
|
||||
|
||||
return result
|
||||
|
||||
except Exception as e:
|
||||
return f"❌ Error searching files: {str(e)}"
|
||||
|
||||
def _find_file(self, filename: str) -> str:
|
||||
"""Find file in uploads directory"""
|
||||
# Clean up filename
|
||||
filename = filename.strip().strip('"').strip("'")
|
||||
|
||||
# Try exact path
|
||||
if os.path.exists(filename):
|
||||
return filename
|
||||
|
||||
# Try in uploads directory with exact name
|
||||
path = os.path.join(self.uploads_dir, filename)
|
||||
if os.path.exists(path):
|
||||
return path
|
||||
|
||||
# Try host mount path
|
||||
path = os.path.join(self.host_uploads_dir, filename)
|
||||
if os.path.exists(path):
|
||||
return path
|
||||
|
||||
# Try glob patterns
|
||||
for directory in [self.uploads_dir, self.host_uploads_dir]:
|
||||
if not os.path.exists(directory):
|
||||
continue
|
||||
|
||||
# UUID prefix match
|
||||
pattern = os.path.join(directory, f"{filename}*")
|
||||
matches = glob.glob(pattern)
|
||||
if matches:
|
||||
return matches[0]
|
||||
|
||||
# Filename contains match
|
||||
pattern = os.path.join(directory, f"*{filename}*")
|
||||
matches = glob.glob(pattern)
|
||||
if matches:
|
||||
# Prefer exact matches
|
||||
for m in matches:
|
||||
if filename in os.path.basename(m):
|
||||
return m
|
||||
return matches[0]
|
||||
|
||||
return None
|
||||
|
||||
def _get_mime_type(self, file_path: str) -> str:
|
||||
"""Determine MIME type from file extension"""
|
||||
ext = os.path.splitext(file_path)[1].lower()
|
||||
mime_types = {
|
||||
'.pdf': 'application/pdf',
|
||||
'.jpg': 'image/jpeg',
|
||||
'.jpeg': 'image/jpeg',
|
||||
'.png': 'image/png',
|
||||
'.gif': 'image/gif',
|
||||
'.tiff': 'image/tiff',
|
||||
'.tif': 'image/tiff',
|
||||
'.bmp': 'image/bmp',
|
||||
'.webp': 'image/webp',
|
||||
'.doc': 'application/msword',
|
||||
'.docx': 'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
|
||||
'.xls': 'application/vnd.ms-excel',
|
||||
'.xlsx': 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
|
||||
'.ppt': 'application/vnd.ms-powerpoint',
|
||||
'.pptx': 'application/vnd.openxmlformats-officedocument.presentationml.presentation',
|
||||
'.epub': 'application/epub+zip',
|
||||
'.txt': 'text/plain',
|
||||
'.html': 'text/html',
|
||||
'.htm': 'text/html',
|
||||
}
|
||||
return mime_types.get(ext, 'application/octet-stream')
|
||||
|
||||
def _clean_content(self, content: str) -> str:
|
||||
"""Clean up extracted content"""
|
||||
if not content:
|
||||
return ""
|
||||
|
||||
# Remove excessive whitespace
|
||||
lines = content.split('\n')
|
||||
cleaned_lines = []
|
||||
prev_empty = False
|
||||
|
||||
for line in lines:
|
||||
stripped = line.strip()
|
||||
if not stripped:
|
||||
if not prev_empty:
|
||||
cleaned_lines.append('')
|
||||
prev_empty = True
|
||||
else:
|
||||
cleaned_lines.append(stripped)
|
||||
prev_empty = False
|
||||
|
||||
# Join and limit consecutive empty lines
|
||||
result = '\n'.join(cleaned_lines)
|
||||
|
||||
# Remove very long lines (>200 chars wrap)
|
||||
if len(result) > 500:
|
||||
# Check if content looks garbled (lots of short fragments)
|
||||
short_words = sum(1 for w in result.split() if len(w) <= 3)
|
||||
total_words = len(result.split())
|
||||
if total_words > 50 and short_words / total_words > 0.3:
|
||||
result += "\n\n⚠️ Warning: This text may contain OCR errors due to complex formatting or handwriting."
|
||||
|
||||
return result
|
||||
|
||||
def _format_size(self, size: int) -> str:
|
||||
"""Format file size"""
|
||||
if size < 1024:
|
||||
return f"{size} B"
|
||||
elif size < 1024 * 1024:
|
||||
return f"{size / 1024:.1f} KB"
|
||||
else:
|
||||
return f"{size / (1024 * 1024):.1f} MB"
|
||||
|
||||
|
||||
# OpenWebUI Tool Functions
|
||||
def extract_file(filename: str) -> str:
|
||||
"""
|
||||
Extract text from a file using Kreuzberg OCR.
|
||||
|
||||
Args:
|
||||
filename: The filename or UUID from the chat (e.g., "document.pdf", "abc123_document.pdf")
|
||||
|
||||
Returns:
|
||||
Extracted text with markdown formatting
|
||||
"""
|
||||
tools = Tools()
|
||||
return tools.extract_file(filename)
|
||||
|
||||
def list_recent_files(count: int = 10) -> str:
|
||||
"""List recent uploaded files"""
|
||||
tools = Tools()
|
||||
return tools.list_recent_files(count)
|
||||
|
||||
def search_files(pattern: str) -> str:
|
||||
"""Search for files by name pattern"""
|
||||
tools = Tools()
|
||||
return tools.search_files(pattern)
|
||||
```
|
||||
|
||||
## Usage Examples
|
||||
|
||||
After installing and activating the tool:
|
||||
|
||||
### Extract text from a file:
|
||||
```
|
||||
extract_file("document.pdf")
|
||||
extract_file("9af3d665-998b-4577-b893-340cb1c28385_document.pdf")
|
||||
extract_file("image.jpg")
|
||||
```
|
||||
|
||||
### List recent files:
|
||||
```
|
||||
list_recent_files()
|
||||
list_recent_files(20)
|
||||
```
|
||||
|
||||
### Search files:
|
||||
```
|
||||
search_files("dnd")
|
||||
search_files("polz")
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
*Updated 11 April 2026 - Added image preprocessing (grayscale + contrast) for better OCR quality*
|
||||
Reference in New Issue
Block a user