from fastapi import FastAPI, UploadFile, File import fitz # PyMuPDF import easyocr from pdf2image import convert_from_bytes import pytesseract import numpy as np app = FastAPI(title="High Accuracy Hindi + English PDF Text Extraction API") # Linux Tesseract path pytesseract.pytesseract.tesseract_cmd = '/usr/bin/tesseract' # Initialize EasyOCR easy_reader = easyocr.Reader(['en', 'hi'], gpu=False) # Root route to test server @app.get("/") def home(): return {"message": "OCR API is running!"} # OCR Endpoint @app.post("/extract_text") async def extract_text(file: UploadFile = File(...)): pdf_bytes = await file.read() text = "" try: # Step 1: Text-based extraction with fitz.open(stream=pdf_bytes, filetype="pdf") as pdf: for page in pdf: text += page.get_text("text") + "\n\n" # Step 2: EasyOCR fallback if len(text.strip()) < 40: with fitz.open(stream=pdf_bytes, filetype="pdf") as pdf: for i, page in enumerate(pdf): pix = page.get_pixmap(dpi=150) img_bytes = pix.tobytes("png") results = easy_reader.readtext(img_bytes, detail=0, paragraph=True) text += " ".join(results) # Step 3: Tesseract fallback for Hindi if len(text.strip()) < 40: images = convert_from_bytes(pdf_bytes, dpi=300) for img in images: text += pytesseract.image_to_string(np.array(img), lang="hin+eng") return {"status": "success", "text": text[:10000]} except Exception as e: return {"status": "error", "message": str(e)}