generated-bill / backend /src /python /ocr_server.py
narasimha01tm's picture
Fix PaddleOCR args and add missing mockBackend files
52d0aa4
Raw History Blame Contribute Delete
9.64 kB
import logging
import sys
import numpy as np
import cv2
import base64
from flask import Flask, request, jsonify
# Configure logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
handlers=[
logging.StreamHandler(sys.stdout)
]
)
logger = logging.getLogger(__name__)
app = Flask(__name__)
# Abstract OCR interface to handle multiple engines
ocr_engine = None
engine_type = None
def init_ocr():
global ocr_engine, engine_type
# Priority 1: PaddleOCR (Best quality)
try:
from paddleocr import PaddleOCR
logger.info("Initializing PaddleOCR model...")
# use_angle_cls=True allows orientation correction
ocr_engine = PaddleOCR(use_textline_orientation=True, lang='en')
engine_type = 'paddle'
logger.info("PaddleOCR model initialized successfully.")
return
except ImportError:
logger.warning("PaddleOCR not found. Falling back to EasyOCR.")
except Exception as e:
logger.warning(f"Failed to initialize PaddleOCR: {str(e)}. Falling back to EasyOCR.")
# Priority 2: EasyOCR (Good fallback)
try:
import easyocr
logger.info("Initializing EasyOCR model...")
# gpu=False for compatibility, True if CUDA available
ocr_engine = easyocr.Reader(['en'], gpu=False)
engine_type = 'easyocr'
logger.info("EasyOCR model initialized successfully.")
return
except ImportError:
logger.error("EasyOCR not found.")
except Exception as e:
logger.error(f"Failed to initialize EasyOCR: {str(e)}")
if ocr_engine is None:
logger.error("No OCR engine could be initialized. Please install paddleocr or easyocr.")
# We don't exit here so the server can at least respond with health checks or errors
def enhance_image(img):
"""
Apply image processing to improve OCR results
"""
try:
# 1. Convert to grayscale if it's color
if len(img.shape) == 3:
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
else:
gray = img
# 2. Rescale image if it's too small (OCR likes larger text)
height, width = gray.shape
if height < 1000 or width < 1000:
scale_factor = 2.0
gray = cv2.resize(gray, None, fx=scale_factor, fy=scale_factor, interpolation=cv2.INTER_CUBIC)
# 3. Denoising
denoised = cv2.fastNlMeansDenoising(gray, None, 10, 7, 21)
# 4. Sharpening using Laplacian
kernel = np.array([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]])
sharpened = cv2.filter2D(denoised, -1, kernel)
# 5. Contrast Limited Adaptive Histogram Equalization (CLAHE)
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
enhanced = clahe.apply(sharpened)
return enhanced
except Exception as e:
logger.warning(f"Image enhancement failed, using original: {str(e)}")
return img
def process_with_paddle(img):
# PaddleOCR returns a list of results per image (list of lists)
result = ocr_engine.ocr(img, cls=True)
structured_data = []
if result and result[0]:
detections = result[0]
for line in detections:
bbox = line[0] # [[x1,y1], [x2,y2], [x3,y3], [x4,y4]]
text_info = line[1] # (text, confidence)
text = text_info[0]
confidence = text_info[1]
# Transform bbox to list of [x, y]
box = [[float(pt[0]), float(pt[1])] for pt in bbox]
h = abs(bbox[0][1] - bbox[2][1])
structured_data.append({
"text": text,
"confidence": float(confidence),
"box": box,
"height": h,
"center_y": (bbox[0][1] + bbox[2][1]) / 2
})
return structured_data
def process_with_easyocr(img):
result = ocr_engine.readtext(img)
structured_data = []
for item in result:
bbox, text, confidence = item
# bbox is [[x1,y1], [x2,y2], [x3,y3], [x4,y4]]
h = abs(bbox[0][1] - bbox[2][1])
box = [[float(pt[0]), float(pt[1])] for pt in bbox]
structured_data.append({
"text": text,
"confidence": float(confidence),
"box": box,
"height": h,
"center_y": (bbox[0][1] + bbox[2][1]) / 2
})
return structured_data
def process_image(image_base64):
try:
if ocr_engine is None:
return {"success": False, "error": "OCR engine not initialized"}
# Decode base64 image
if ',' in image_base64:
image_base64 = image_base64.split(',')[1]
img_bytes = base64.b64decode(image_base64)
nparr = np.frombuffer(img_bytes, np.uint8)
img = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
if img is None:
return {"success": False, "error": "Failed to decode image"}
# Apply enhancement
processed_img = enhance_image(img)
# Perform OCR based on engine
structured_data = []
if engine_type == 'paddle':
structured_data = process_with_paddle(processed_img)
elif engine_type == 'easyocr':
structured_data = process_with_easyocr(processed_img)
# Reconstruct text with line grouping logic
# Calculate average text height for dynamic thresholding
total_height = sum(item['height'] for item in structured_data) if structured_data else 0
count = len(structured_data)
# Dynamic y_threshold based on average line height
avg_height = (total_height / count) if count > 0 else 20
y_threshold = avg_height * 0.5 # 50% of avg height as same-line threshold
# Sort by Y-coordinate
structured_data.sort(key=lambda x: x['box'][0][1])
lines = []
if structured_data:
current_line = []
current_y = -1
for item in structured_data:
y_center = item['center_y']
if current_y == -1:
current_y = y_center
current_line.append(item)
elif abs(y_center - current_y) < y_threshold:
current_line.append(item)
else:
# Sort current line by X-coordinate
current_line.sort(key=lambda x: x['box'][0][0])
lines.append(" ".join([str(i.get('text', '')) for i in current_line]))
current_line = [item]
current_y = y_center
if current_line:
current_line.sort(key=lambda x: x['box'][0][0])
lines.append(" ".join([str(i.get('text', '')) for i in current_line]))
# Build full text from lines with enhancement (User request)
import re
final_lines = []
for line_str in lines:
# 1. Filter out "thank you" and "visit" phrases
junk_keywords = ["thank you", "visit again", "thanks", "welcome", "have a nice day", "visit us", "come again"]
if any(k in line_str.lower() for k in junk_keywords):
continue
# 2. Remove any numbers (User request: "dont add any number")
# This helps in cleaning up item names for real-time price fetching
cleaned = re.sub(r'\d+', '', line_str)
# Clean up extra whitespace and punctuation
cleaned = re.sub(r'\s+', ' ', cleaned).strip()
# Skip empty lines or lines with no alphanumeric characters
if cleaned and any(c.isalnum() for c in cleaned):
final_lines.append(cleaned)
reconstructed_text = "\n".join(final_lines)
return {
"success": True,
"text": reconstructed_text,
"structured_data": [
{k: v for k, v in item.items() if k != 'center_y' and k != 'height'}
for item in structured_data
],
"engine": engine_type
}
except Exception as e:
logger.error(f"Error processing image: {str(e)}")
import traceback
traceback.print_exc()
return {"success": False, "error": str(e)}
@app.route('/ocr', methods=['POST'])
def ocr_endpoint():
try:
data = request.get_json()
if not data or 'imageBase64' not in data:
return jsonify({"success": False, "error": "No imageBase64 provided"}), 400
image_base64 = data['imageBase64']
# Ensure OCR is ready
if ocr_engine is None:
init_ocr()
result = process_image(image_base64)
return jsonify(result)
except Exception as e:
logger.error(f"Endpoint error: {str(e)}")
return jsonify({"success": False, "error": str(e)}), 500
@app.route('/health', methods=['GET'])
def health_check():
status = "healthy" if ocr_engine is not None else "initializing_or_failed"
return jsonify({
"status": status,
"model": engine_type or "none",
"enhanced": True
})
if __name__ == '__main__':
# Initialize implementation on startup
init_ocr()
port = 5001
logger.info(f"Starting Enhanced OCR Server ({engine_type or 'No Engine'}) on port {port}")
app.run(host='0.0.0.0', port=port)