feat(ocr): add self-hosted OCR engine support (Tesseract, EasyOCR) with multi-provider cross-checking
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot]
@@ -17,6 +17,7 @@ from app.tasks.extract_metadata_with_gpt import extract_metadata_with_gpt
|
||||
from app.tasks.process_with_azure_document_intelligence import (
|
||||
process_with_azure_document_intelligence,
|
||||
)
|
||||
from app.tasks.process_with_ocr import process_with_ocr
|
||||
from app.tasks.retry_config import BaseTaskWithRetry
|
||||
from app.utils import get_unique_filepath_with_counter, hash_file, log_task_progress
|
||||
|
||||
@@ -313,7 +314,7 @@ def process_document(
|
||||
"Queued for forced OCR processing",
|
||||
file_id=file_id,
|
||||
)
|
||||
process_with_azure_document_intelligence.delay(new_filename, file_id)
|
||||
process_with_ocr.delay(new_filename, file_id)
|
||||
return {"file": new_local_path, "status": "Queued for forced OCR", "file_id": file_id}
|
||||
|
||||
# If the file is not a PDF, skip embedded text check and convert to PDF first
|
||||
@@ -411,12 +412,12 @@ def process_document(
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
# Mark Azure OCR as skipped since we extracted text locally
|
||||
# Mark OCR as skipped since we extracted text locally
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"process_with_azure_document_intelligence",
|
||||
"skipped",
|
||||
"Local text extraction succeeded, Azure OCR not needed",
|
||||
"Local text extraction succeeded, OCR not needed",
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
@@ -436,8 +437,8 @@ def process_document(
|
||||
"file_id": file_id,
|
||||
}
|
||||
|
||||
# 3. If no embedded text, queue Azure Document Intelligence processing
|
||||
logger.info(f"[{task_id}] No embedded text found. Queueing Azure Document Intelligence processing")
|
||||
# 3. If no embedded text, queue OCR processing
|
||||
logger.info(f"[{task_id}] No embedded text found. Queueing OCR processing")
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"check_text",
|
||||
@@ -446,12 +447,12 @@ def process_document(
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
# Mark local text extraction as skipped since we're using Azure OCR
|
||||
# Mark local text extraction as skipped since we're using cloud OCR
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"extract_text",
|
||||
"skipped",
|
||||
"No embedded text, using Azure OCR instead",
|
||||
"No embedded text, using OCR instead",
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
@@ -462,5 +463,5 @@ def process_document(
|
||||
"Queued for OCR processing",
|
||||
file_id=file_id,
|
||||
)
|
||||
process_with_azure_document_intelligence.delay(new_filename, file_id)
|
||||
process_with_ocr.delay(new_filename, file_id)
|
||||
return {"file": new_local_path, "status": "Queued for OCR", "file_id": file_id}
|
||||
|
||||
@@ -0,0 +1,139 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Unified OCR processing task for DocuElevate.
|
||||
|
||||
This task replaces the single-provider ``process_with_azure_document_intelligence``
|
||||
task with a multi-engine OCR pipeline that:
|
||||
|
||||
1. Runs every OCR provider listed in ``OCR_PROVIDERS`` (default: ``azure``).
|
||||
2. Merges/cross-checks the results using the configured AI model when more
|
||||
than one provider is active (see ``OCR_MERGE_STRATEGY``).
|
||||
3. Writes the best searchable PDF back to the working directory.
|
||||
4. Hands off to the page-rotation and metadata-extraction pipeline exactly as
|
||||
the legacy Azure task did.
|
||||
"""
|
||||
|
||||
import logging
|
||||
import os
|
||||
from typing import Optional
|
||||
|
||||
from app.celery_app import celery
|
||||
from app.config import settings
|
||||
from app.tasks.retry_config import BaseTaskWithRetry
|
||||
from app.tasks.rotate_pdf_pages import rotate_pdf_pages
|
||||
from app.utils import log_task_progress
|
||||
from app.utils.ocr_provider import OCRResult, get_ocr_providers, merge_ocr_results
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
@celery.task(base=BaseTaskWithRetry, bind=True)
|
||||
def process_with_ocr(self, filename: str, file_id: Optional[int] = None):
|
||||
"""Run the configured OCR providers on *filename* and continue the pipeline.
|
||||
|
||||
When multiple OCR providers are configured the results are merged using the
|
||||
AI model (or a simpler strategy controlled by ``OCR_MERGE_STRATEGY``).
|
||||
|
||||
Args:
|
||||
filename: Base name of the file inside ``<workdir>/tmp/``.
|
||||
file_id: Optional database record ID passed through to downstream tasks.
|
||||
"""
|
||||
task_id = self.request.id
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"process_with_ocr",
|
||||
"in_progress",
|
||||
f"Starting OCR for {filename}",
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
try:
|
||||
tmp_file_path = os.path.join(settings.workdir, "tmp", filename)
|
||||
if not os.path.exists(tmp_file_path):
|
||||
raise FileNotFoundError(f"Local file not found: {tmp_file_path}")
|
||||
|
||||
providers = get_ocr_providers()
|
||||
provider_names = [p.name for p in providers]
|
||||
logger.info(f"[{task_id}] Running {len(providers)} OCR provider(s): {provider_names}")
|
||||
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"run_ocr_providers",
|
||||
"in_progress",
|
||||
f"Running OCR providers: {', '.join(provider_names)}",
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
results = []
|
||||
errors = []
|
||||
for provider in providers:
|
||||
pname = provider.__class__.__name__
|
||||
try:
|
||||
logger.info(f"[{task_id}] Running {pname} on {filename}")
|
||||
result: OCRResult = provider.process(tmp_file_path)
|
||||
results.append(result)
|
||||
logger.info(f"[{task_id}] {pname} extracted {len(result.text)} chars")
|
||||
except Exception as exc:
|
||||
logger.error(f"[{task_id}] {pname} failed for {filename}: {exc}")
|
||||
errors.append(f"{pname}: {exc}")
|
||||
|
||||
if not results:
|
||||
error_summary = "; ".join(errors)
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"run_ocr_providers",
|
||||
"failure",
|
||||
"All OCR providers failed",
|
||||
file_id=file_id,
|
||||
detail=error_summary,
|
||||
)
|
||||
raise RuntimeError(f"All OCR providers failed for {filename}: {error_summary}")
|
||||
|
||||
if errors:
|
||||
logger.warning(f"[{task_id}] Some OCR providers failed: {'; '.join(errors)}")
|
||||
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"run_ocr_providers",
|
||||
"success",
|
||||
f"{len(results)} of {len(providers)} OCR provider(s) succeeded",
|
||||
file_id=file_id,
|
||||
)
|
||||
|
||||
# Merge results (no-op when only one provider succeeded)
|
||||
extracted_text, searchable_pdf_path, rotation_data = merge_ocr_results(results, filename)
|
||||
logger.info(
|
||||
f"[{task_id}] Merged OCR text: {len(extracted_text)} chars, "
|
||||
f"pdf={'yes' if searchable_pdf_path else 'no'}, "
|
||||
f"rotations={len(rotation_data)}"
|
||||
)
|
||||
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"process_with_ocr",
|
||||
"success",
|
||||
f"OCR complete for {filename}",
|
||||
file_id=file_id,
|
||||
detail=f"Extracted {len(extracted_text)} chars using {len(results)} provider(s)",
|
||||
)
|
||||
|
||||
# Continue pipeline: rotate pages (if needed), then extract metadata
|
||||
rotate_pdf_pages.delay(filename, extracted_text, rotation_data, file_id)
|
||||
|
||||
return {
|
||||
"file": filename,
|
||||
"searchable_pdf": searchable_pdf_path or tmp_file_path,
|
||||
"cleaned_text": extracted_text,
|
||||
"providers_used": [r.provider for r in results],
|
||||
}
|
||||
|
||||
except Exception as exc:
|
||||
logger.error(f"[{task_id}] OCR failed for {filename}: {exc}")
|
||||
log_task_progress(
|
||||
task_id,
|
||||
"process_with_ocr",
|
||||
"failure",
|
||||
f"OCR failed for {filename}",
|
||||
file_id=file_id,
|
||||
detail=str(exc),
|
||||
)
|
||||
raise
|
||||
Reference in New Issue
Block a user