f064b27439
Kurrent OCR produces much lower confidence than typewriter/Latin. Separate thresholds allow aggressive filtering for Kurrent (0.5) while keeping typewriter lenient (0.3). - OCR_CONFIDENCE_THRESHOLD: default for Surya paths (0.3) - OCR_CONFIDENCE_THRESHOLD_KURRENT: Kraken Kurrent path (0.5) - apply_confidence_markers() now accepts threshold parameter - get_threshold(script_type) selects the right threshold Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
101 lines
3.1 KiB
Python
101 lines
3.1 KiB
Python
"""OCR microservice — FastAPI app with Surya and Kraken engine support."""
|
|
|
|
import io
|
|
import logging
|
|
from contextlib import asynccontextmanager
|
|
|
|
import httpx
|
|
import pypdfium2 as pdfium
|
|
from fastapi import FastAPI, HTTPException
|
|
from PIL import Image
|
|
|
|
from confidence import apply_confidence_markers, get_threshold
|
|
from engines import kraken as kraken_engine
|
|
from engines import surya as surya_engine
|
|
from models import OcrBlock, OcrRequest
|
|
|
|
logging.basicConfig(level=logging.INFO)
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_models_ready = False
|
|
|
|
|
|
@asynccontextmanager
|
|
async def lifespan(app: FastAPI):
|
|
"""Load all OCR models at startup before accepting requests."""
|
|
global _models_ready
|
|
|
|
logger.info("Loading OCR models at startup...")
|
|
surya_engine.load_models()
|
|
kraken_engine.load_models()
|
|
_models_ready = True
|
|
logger.info("All OCR models loaded — ready to accept requests")
|
|
|
|
yield
|
|
|
|
logger.info("Shutting down OCR service")
|
|
|
|
|
|
app = FastAPI(title="Familienarchiv OCR Service", lifespan=lifespan)
|
|
|
|
|
|
@app.get("/health")
|
|
def health():
|
|
"""Health endpoint — returns 200 only after models are loaded."""
|
|
if not _models_ready:
|
|
raise HTTPException(status_code=503, detail="Models not loaded yet")
|
|
return {"status": "ok", "surya": True, "kraken": kraken_engine.is_available()}
|
|
|
|
|
|
@app.post("/ocr", response_model=list[OcrBlock])
|
|
async def run_ocr(request: OcrRequest):
|
|
"""Run OCR on a PDF document.
|
|
|
|
Downloads the PDF from the provided URL, converts pages to images,
|
|
and runs the appropriate OCR engine based on scriptType.
|
|
"""
|
|
if not _models_ready:
|
|
raise HTTPException(status_code=503, detail="Models not loaded yet")
|
|
|
|
images = await _download_and_convert_pdf(request.pdf_url)
|
|
|
|
script_type = request.script_type.upper()
|
|
|
|
if script_type == "HANDWRITING_KURRENT":
|
|
if not kraken_engine.is_available():
|
|
raise HTTPException(
|
|
status_code=400,
|
|
detail="Kraken model not available — cannot process Kurrent script",
|
|
)
|
|
blocks = kraken_engine.extract_blocks(images, request.language)
|
|
else:
|
|
# TYPEWRITER, HANDWRITING_LATIN, UNKNOWN — all use Surya
|
|
blocks = surya_engine.extract_blocks(images, request.language)
|
|
|
|
threshold = get_threshold(script_type)
|
|
for block in blocks:
|
|
if block.get("words"):
|
|
block["text"] = apply_confidence_markers(block["words"], threshold)
|
|
block.pop("words", None)
|
|
|
|
return [OcrBlock(**b) for b in blocks]
|
|
|
|
|
|
async def _download_and_convert_pdf(url: str) -> list[Image.Image]:
|
|
"""Download a PDF from URL and convert each page to a PIL Image."""
|
|
async with httpx.AsyncClient(timeout=httpx.Timeout(300.0)) as client:
|
|
response = await client.get(url)
|
|
response.raise_for_status()
|
|
|
|
pdf = pdfium.PdfDocument(io.BytesIO(response.content))
|
|
images = []
|
|
|
|
for page_idx in range(len(pdf)):
|
|
page = pdf[page_idx]
|
|
# Render at 300 DPI for good OCR quality
|
|
bitmap = page.render(scale=300 / 72)
|
|
pil_image = bitmap.to_pil()
|
|
images.append(pil_image)
|
|
|
|
return images
|