pdf-ocr / app.py
ChintanSatva's picture
Update app.py
863243a verified
from fastapi import FastAPI, File, UploadFile, HTTPException
from decimal import Decimal, InvalidOperation
from fastapi.encoders import jsonable_encoder
from starlette.responses import JSONResponse
import pytesseract
import cv2
import os
from PIL import Image
import json
import unicodedata
from pdf2image import convert_from_bytes
from pypdf import PdfReader
import numpy as np
from typing import List, Any
import io
import logging
import time
import asyncio
import psutil
import cachetools
import hashlib
import re
import google.generativeai as genai
from dotenv import load_dotenv
# --- START OF MODIFICATIONS ---
# 1. Define a custom JSON encoder function
def custom_encoder(obj: Any) -> Any:
if isinstance(obj, Decimal):
try:
float_val = float(obj)
if float_val == 0:
return "0.0"
elif 0 < abs(float_val) < 1e-10:
result = f"{float_val:.20f}".rstrip('0').rstrip('.')
elif 0 < abs(float_val) < 1e-6:
result = f"{float_val:.15f}".rstrip('0').rstrip('.')
elif abs(float_val) < 1:
result = f"{float_val:.10f}".rstrip('0').rstrip('.')
else:
result = f"{float_val:.8f}".rstrip('0').rstrip('.')
# Ensure the result is a string to prevent JSON serialization issues
return str(result)
except (ValueError, OverflowError, InvalidOperation):
return str(obj) # Fallback to string representation
return jsonable_encoder(obj)
def custom_decimal_parser(s):
"""
Custom parser that ensures numbers are converted to decimal format.
"""
try:
float_val = float(s)
if float_val == 0:
return Decimal('0.0')
elif 0 < abs(float_val) < 1e-10:
formatted = f"{float_val:.20f}".rstrip('0').rstrip('.')
elif 0 < abs(float_val) < 1e-6:
formatted = f"{float_val:.15f}".rstrip('0').rstrip('.')
elif abs(float_val) < 1:
formatted = f"{float_val:.10f}".rstrip('0').rstrip('.')
else:
formatted = f"{float_val:.8f}".rstrip('0').rstrip('.')
return Decimal(formatted)
except (ValueError, InvalidOperation):
return Decimal(str(s))
def fix_scientific_notation_in_json(json_str):
"""
Fix scientific notation in JSON string before parsing.
"""
def replace_scientific(match):
try:
scientific_num = match.group(0)
float_val = float(scientific_num)
if float_val == 0:
return "0.0"
elif 0 < abs(float_val) < 1e-10:
return f"{float_val:.20f}".rstrip('0').rstrip('.') or "0.0"
elif 0 < abs(float_val) < 1e-6:
return f"{float_val:.15f}".rstrip('0').rstrip('.') or "0.0"
elif abs(float_val) < 1:
return f"{float_val:.10f}".rstrip('0').rstrip('.') or "0.0"
else:
return f"{float_val:.8f}".rstrip('0').rstrip('.') or "0.0"
except Exception as e:
logger.error(f"Error converting {match.group(0)}: {e}")
return match.group(0)
patterns = [
r'-?\d+\.?\d*[eE][+-]?\d+',
r'-?\d+[eE][+-]?\d+',
r'-?\d+\.\d+[eE][+-]?\d+',
]
original_json = json_str
for pattern in patterns:
json_str = re.sub(pattern, replace_scientific, json_str)
def replace_quoted_scientific(match):
full_match = match.group(0)
number_part = match.group(1)
try:
float_val = float(number_part)
if 0 < abs(float_val) < 1e-6:
converted = f"{float_val:.15f}".rstrip('0').rstrip('.') or "0.0"
else:
converted = f"{float_val:.10f}".rstrip('0').rstrip('.') or "0.0"
return f'"{converted}"'
except:
return full_match
quoted_pattern = r'"(-?\d+\.?\d*[eE][+-]?\d+)"'
json_str = re.sub(quoted_pattern, replace_quoted_scientific, json_str)
if original_json != json_str:
logger.info(f"JSON transformation occurred")
logger.info(f"Original: {original_json[:200]}...")
logger.info(f"Fixed: {json_str[:200]}...")
return json_str
def convert_scientific_decimals(obj):
"""
Recursively convert Decimal objects to proper decimal notation.
"""
if isinstance(obj, dict):
return {k: convert_scientific_decimals(v) for k, v in obj.items()}
elif isinstance(obj, list):
return [convert_scientific_decimals(item) for item in obj]
elif isinstance(obj, Decimal):
try:
float_val = float(obj)
if float_val == 0:
return Decimal('0.0')
elif 0 < abs(float_val) < 1e-10:
formatted = f"{float_val:.20f}".rstrip('0').rstrip('.')
elif 0 < abs(float_val) < 1e-6:
formatted = f"{float_val:.15f}".rstrip('0').rstrip('.')
elif abs(float_val) < 1:
formatted = f"{float_val:.10f}".rstrip('0').rstrip('.')
elif abs(float_val) < 1000000:
formatted = f"{float_val:.8f}".rstrip('0').rstrip('.')
else:
formatted = str(int(float_val)) if float_val == int(float_val) else f"{float_val:.2f}".rstrip('0').rstrip('.')
if formatted == '0' and float_val != 0:
formatted = f"{float_val:.20f}".rstrip('0').rstrip('.')
return Decimal(formatted)
except (ValueError, OverflowError, InvalidOperation):
return obj
else:
return obj
def force_decimal_format(data):
"""
Ensure all numeric values are in proper decimal format before JSON encoding.
"""
if isinstance(data, dict):
result = {}
for key, value in data.items():
if key in ['unit_price', 'total_price', 'tax_amount', 'discount', 'net_amount',
'sub_total', 'tax_total', 'discount_total', 'total_amount', 'tax_rate']:
if isinstance(value, dict) and 'value' in value:
if isinstance(value['value'], (Decimal, float, int)):
try:
float_val = float(value['value'])
if float_val == 0:
decimal_str = "0.0"
elif 0 < abs(float_val) < 1e-10:
decimal_str = f"{float_val:.20f}".rstrip('0').rstrip('.')
elif 0 < abs(float_val) < 1e-6:
decimal_str = f"{float_val:.15f}".rstrip('0').rstrip('.')
else:
decimal_str = f"{float_val:.10f}".rstrip('0').rstrip('.')
result[key] = {'value': Decimal(decimal_str), 'accuracy': value['accuracy']}
except (ValueError, InvalidOperation):
result[key] = value
else:
result[key] = value
else:
result[key] = force_decimal_format(value)
else:
result[key] = force_decimal_format(value)
return result
elif isinstance(data, list):
return [force_decimal_format(item) for item in data]
else:
return data
# --- END OF MODIFICATIONS ---
app = FastAPI()
# Configure logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
# Load environment variables
load_dotenv()
# Configure Gemini API
api_key = os.getenv("GOOGLE_API_KEY")
if not api_key:
logger.error("GOOGLE_API_KEY not set")
raise HTTPException(status_code=500, detail="GOOGLE_API_KEY not set")
genai.configure(api_key=api_key)
model = genai.GenerativeModel("gemini-2.5-flash")
# Set Tesseract path
pytesseract.pytesseract.tesseract_cmd = "/usr/bin/tesseract"
# In-memory caches
raw_text_cache = cachetools.TTLCache(maxsize=100, ttl=3600)
structured_data_cache = cachetools.TTLCache(maxsize=100, ttl=3600)
def log_memory_usage():
"""Log current memory usage."""
process = psutil.Process()
mem_info = process.memory_info()
return f"Memory usage: {mem_info.rss / 1024 / 1024:.2f} MB"
def get_file_hash(file_bytes):
"""Generate MD5 hash of file content."""
return hashlib.md5(file_bytes).hexdigest()
def get_text_hash(raw_text):
"""Generate MD5 hash of raw text."""
return hashlib.md5(raw_text.encode('utf-8')).hexdigest()
async def process_image(img_bytes, filename, idx):
"""Process a single image with OCR."""
start_time = time.time()
logger.info(f"Starting OCR for {filename} image {idx}, {log_memory_usage()}")
try:
img = Image.open(io.BytesIO(img_bytes))
img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
gray = cv2.cvtColor(img_cv, cv2.COLOR_BGR2GRAY)
img_pil = Image.fromarray(cv2.cvtColor(gray, cv2.COLOR_GRAY2RGB))
custom_config = r'--oem 1 --psm 6 -l eng+ara'
page_text = pytesseract.image_to_string(img_pil, config=custom_config)
logger.info(f"Completed OCR for {filename} image {idx}, took {time.time() - start_time:.2f} seconds, {log_memory_usage()}")
return page_text + "\n"
except Exception as e:
logger.error(f"OCR failed for {filename} image {idx}: {str(e)}, {log_memory_usage()}")
return ""
async def process_pdf_page(img, page_idx):
"""Process a single PDF page with OCR."""
start_time = time.time()
logger.info(f"Starting OCR for PDF page {page_idx}, {log_memory_usage()}")
try:
img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
gray = cv2.cvtColor(img_cv, cv2.COLOR_BGR2GRAY)
img_pil = Image.fromarray(cv2.cvtColor(gray, cv2.COLOR_GRAY2RGB))
custom_config = r'--oem 1 --psm 6 -l eng+ara'
page_text = pytesseract.image_to_string(img_pil, config=custom_config)
logger.info(f"Completed OCR for PDF page {page_idx}, took {time.time() - start_time:.2f} seconds, {log_memory_usage()}")
return page_text + "\n"
except Exception as e:
logger.error(f"OCR failed for PDF page {page_idx}: {str(e)}, {log_memory_usage()}")
return ""
async def process_with_gemini(filename: str, raw_text: str):
"""Process raw text with Gemini to extract structured data."""
start_time = time.time()
logger.info(f"Starting Gemini processing for {filename}, {log_memory_usage()}")
text_hash = get_text_hash(raw_text)
if text_hash in structured_data_cache:
logger.info(f"Structured data cache hit for {filename}, {log_memory_usage()}")
return structured_data_cache[text_hash]
if len(raw_text) > 20000:
raw_text = raw_text[:20000]
logger.info(f"Truncated raw text for {filename} to 20000 characters, {log_memory_usage()}")
try:
prompt = f"""You are an intelligent invoice data extractor. Given raw text from an invoice (in English or other languages),
extract key business fields into the specified JSON format. Return each field along with an estimated accuracy score between 0 and 1.
- Accuracy reflects your confidence in the correctness of each field.
- Handle synonyms (e.g., 'total' = 'net', 'tax' = 'GST'/'TDS').
- Detect currency from symbols ($, ₹, €) or keywords (USD, INR, EUR); default to USD if unclear.
- The 'items' list may have multiple entries, each with detailed attributes.
- If a field is missing or not found, return an empty value (`""` or `0`) and set `accuracy` to `0.0`.
- Convert any date found in format: YYYY-MM-DD
CRITICAL: ALL numeric values must be in full decimal notation. NEVER EVER use scientific notation or exponential format:
- CORRECT: 0.0000009, 0.00000015, 0.0000002, 1500000, 0.00123
- ABSOLUTELY FORBIDDEN: 9e-7, 9E-7, 1.5e-7, 1.5E-7, 2e-7, 2E-7, 1.5e+6, 1.23e-3, any number with 'e' or 'E'
- For very small numbers like 0.0000009, you MUST write out all the zeros: 0.0000009
- For large numbers like 1500000, you MUST write out all the digits: 1500000
- This is MANDATORY for: unit_price, total_price, tax_amount, discount, net_amount, sub_total, tax_total, discount_total, total_amount
- Example: if unit price is 9 * 10^-7, write it as 0.0000009, NOT 9e-7 or 9E-7
Raw text:
{raw_text}
Output JSON:
{{
"invoice": {{
"invoice_number": {{"value": "", "accuracy": 0.0}},
"invoice_date": {{"value": "", "accuracy": 0.0}},
"due_date": {{"value": "", "accuracy": 0.0}},
"purchase_order_number": {{"value": "", "accuracy": 0.0}},
"vendor": {{
"vendor_id": {{"value": "", "accuracy": 0.0}},
"name": {{"value": "", "accuracy": 0.0}},
"address": {{
"line1": {{"value": "", "accuracy": 0.0}},
"line2": {{"value": "", "accuracy": 0.0}},
"city": {{"value": "", "accuracy": 0.0}},
"state": {{"value": "", "accuracy": 0.0}},
"postal_code": {{"value": "", "accuracy": 0.0}},
"country": {{"value": "", "accuracy": 0.0}}
}},
"contact": {{
"email": {{"value": "", "accuracy": 0.0}},
"phone": {{"value": "", "accuracy": 0.0}}
}},
"tax_id": {{"value": "", "accuracy": 0.0}}
}},
"buyer": {{
"buyer_id": {{"value": "", "accuracy": 0.0}},
"name": {{"value": "", "accuracy": 0.0}},
"address": {{
"line1": {{"value": "", "accuracy": 0.0}},
"line2": {{"value": "", "accuracy": 0.0}},
"city": {{"value": "", "accuracy": 0.0}},
"state": {{"value": "", "accuracy": 0.0}},
"postal_code": {{"value": "", "accuracy": 0.0}},
"country": {{"value": "", "accuracy": 0.0}}
}},
"contact": {{
"email": {{"value": "", "accuracy": 0.0}},
"phone": {{"value": "", "accuracy": 0.0}}
}},
"tax_id": {{"value": "", "accuracy": 0.0}}
}},
"items": [
{{
"item_id": {{"value": "", "accuracy": 0.0}},
"description": {{"value": "", "accuracy": 0.0}},
"quantity": {{"value": 0, "accuracy": 0.0}},
"unit_of_measure": {{"value": "", "accuracy": 0.0}},
"unit_price": {{"value": 0.0, "accuracy": 0.0}},
"total_price": {{"value": 0.0, "accuracy": 0.0}},
"tax_rate": {{"value": 0.0, "accuracy": 0.0}},
"tax_amount": {{"value": 0.0, "accuracy": 0.0}},
"discount": {{"value": 0.0, "accuracy": 0.0}},
"net_amount": {{"value": 0.0, "accuracy": 0.0}}
}}
],
"sub_total": {{"value": 0.0, "accuracy": 0.0}},
"tax_total": {{"value": 0.0, "accuracy": 0.0}},
"discount_total": {{"value": 0.0, "accuracy": 0.0}},
"total_amount": {{"value": 0.0, "accuracy": 0.0}},
"currency": {{"value": "", "accuracy": 0.0}}
}}
}}
"""
response = model.generate_content(prompt)
llm_output = response.text
json_start = llm_output.find("{")
json_end = llm_output.rfind("}") + 1
json_str = llm_output[json_start:json_end]
logger.info(f"Extracted JSON before fix: {json_str}")
json_str = fix_scientific_notation_in_json(json_str)
structured_data = json.loads(json_str, parse_float=custom_decimal_parser)
structured_data = convert_scientific_decimals(structured_data)
structured_data = force_decimal_format(structured_data)
structured_data_cache[text_hash] = structured_data
logger.info(f"Gemini processing for {filename}, took {time.time() - start_time:.2f} seconds, {log_memory_usage()}")
# Log structured data with custom encoder to avoid scientific notation in logs
log_friendly_data = json.dumps(structured_data, default=custom_encoder)
return structured_data
except Exception as e:
logger.error(f"Gemini processing failed for {filename}: {str(e)}, {log_memory_usage()}")
return {"error": f"Gemini processing failed: {str(e)}"}
@app.post("/ocr")
async def extract_and_structure(files: List[UploadFile] = File(...)):
output_data = {
"success": True,
"message": "",
"data": []
}
success_count = 0
fail_count = 0
logger.info(f"Starting processing for {len(files)} files, {log_memory_usage()}")
for file in files:
total_start_time = time.time()
logger.info(f"Processing file: {file.filename}, {log_memory_usage()}")
valid_extensions = {'.pdf', '.jpg', '.jpeg', '.png'}
file_ext = os.path.splitext(file.filename.lower())[1]
if file_ext not in valid_extensions:
fail_count += 1
output_data["data"].append({
"filename": file.filename,
"structured_data": {"error": f"Unsupported file format: {file_ext}"},
"error": f"Unsupported file format: {file_ext}"
})
logger.error(f"Unsupported file format for {file.filename}: {file_ext}")
continue
try:
file_start_time = time.time()
file_bytes = await file.read()
file_stream = io.BytesIO(file_bytes)
file_hash = get_file_hash(file_bytes)
logger.info(f"Read file {file.filename}, took {time.time() - file_start_time:.2f} seconds, size: {len(file_bytes)/1024:.2f} KB, {log_memory_usage()}")
except Exception as e:
fail_count += 1
output_data["data"].append({
"filename": file.filename,
"structured_data": {"error": f"Failed to read file: {str(e)}"},
"error": f"Failed to read file: {str(e)}"
})
logger.error(f"Failed to read file {file.filename}: {str(e)}, {log_memory_usage()}")
continue
raw_text = ""
if file_hash in raw_text_cache:
raw_text = raw_text_cache[file_hash]
logger.info(f"Raw text cache hit for {file.filename}, {log_memory_usage()}")
else:
if file_ext == '.pdf':
try:
extract_start_time = time.time()
reader = PdfReader(file_stream)
for page in reader.pages:
text = page.extract_text()
if text:
raw_text += text + "\n"
logger.info(f"Embedded text extraction for {file.filename}, took {time.time() - extract_start_time:.2f} seconds, text length: {len(raw_text)}, {log_memory_usage()}")
except Exception as e:
logger.warning(f"Embedded text extraction failed for {file.filename}: {str(e)}, {log_memory_usage()}")
if not raw_text.strip():
try:
convert_start_time = time.time()
images = convert_from_bytes(file_bytes, dpi=150)
logger.info(f"PDF to images conversion for {file.filename}, {len(images)} pages, took {time.time() - convert_start_time:.2f} seconds, {log_memory_usage()}")
ocr_tasks = [process_pdf_page(img, i) for i, img in enumerate(images)]
page_texts = await asyncio.gather(*ocr_tasks)
raw_text = "".join(page_texts)
logger.info(f"Total OCR for {file.filename}, text length: {len(raw_text)}, {log_memory_usage()}")
except Exception as e:
fail_count += 1
output_data["data"].append({
"filename": file.filename,
"structured_data": {"error": f"OCR failed: {str(e)}"},
"error": f"OCR failed: {str(e)}"
})
logger.error(f"OCR failed for {file.filename}: {str(e)}, {log_memory_usage()}")
continue
else:
try:
raw_text = await process_image(file_bytes, file.filename, 0)
logger.info(f"Image OCR for {file.filename}, text length: {len(raw_text)}, {log_memory_usage()}")
except Exception as e:
fail_count += 1
output_data["data"].append({
"filename": file.filename,
"structured_data": {"error": f"Image OCR failed: {str(e)}"},
"error": f"Image OCR failed: {str(e)}"
})
logger.error(f"Image OCR failed for {file.filename}: {str(e)}, {log_memory_usage()}")
continue
if raw_text:
raw_text = unicodedata.normalize('NFKC', raw_text)
raw_text_cache[file_hash] = raw_text
structured_data = await process_with_gemini(file.filename, raw_text)
if "error" not in structured_data:
success_count += 1
else:
fail_count += 1
output_data["data"].append({
"filename": file.filename,
"structured_data": structured_data,
"error": structured_data.get("error", "")
})
logger.info(f"Total processing for {file.filename}, took {time.time() - total_start_time:.2f} seconds, {log_memory_usage()}")
output_data["message"] = f"Processed {len(files)} files. {success_count} succeeded, {fail_count} failed."
if fail_count > 0 and success_count == 0:
output_data["success"] = False
logger.info(f"Completed processing for {len(files)} files, {success_count} succeeded, {fail_count} failed, {log_memory_usage()}")
output_data = force_decimal_format(output_data)
encoded_data = json.dumps(output_data, default=custom_encoder)
return JSONResponse(content=json.loads(encoded_data))