Initial commit

utils/pdf.py

@@ -0,0 +1,85 @@
+from pathlib import Path
+
+import fitz  # PyMuPDF
+from loguru import logger
+
+import config
+import concurrent.futures
+
+
+def is_pdf(filename: str) -> bool:
+    return Path(filename).suffix.lower() == ".pdf"
+
+
+def pdf_to_images(pdf_path: str, dpi: int = config.OCR_DPI,
+                  max_pages: int = config.PDF_MAX_PAGES) -> list[Path]:
+    pdf_path   = Path(pdf_path)
+    output_dir = pdf_path.parent / pdf_path.stem
+    output_dir.mkdir(exist_ok=True)
+
+    ext          = config.PDF_IMAGE_FORMAT                     # "jpeg"
+    page_indices = list(range(max_pages))
+
+    # 이미 변환된 이미지가 모두 있으면 스킵
+    cached = [output_dir / f"page_{i+1:03d}.{ext}" for i in page_indices]
+    if all(p.exists() for p in cached):
+        logger.debug(f"PDF 변환 캐시 사용: {pdf_path.name}")
+        return cached
+
+    pdf_path = Path(pdf_path)
+    print(pdf_path.exists())
+    print(pdf_path.stat().st_size)
+    if not pdf_path.exists() or pdf_path.stat().st_size == 0:
+        raise ValueError(f"유효하지 않은 PDF 파일입니다: {pdf_path.name}")
+
+    matrix = fitz.Matrix(dpi / 72, dpi / 72)
+    doc = fitz.open(str(pdf_path))
+
+    img_paths = []
+    for i in range(min(max_pages, len(doc))):
+        img_path = output_dir / f"page_{i+1:03d}.{ext}"
+        if img_path.exists():
+            logger.debug(f"페이지 캐시 사용: {img_path.name}")
+        else:
+            pixmap = doc[i].get_pixmap(matrix=matrix, alpha=False)
+
+            # 최대 너비 초과 시 비율 유지하며 축소
+            if hasattr(config, 'PDF_MAX_WIDTH') and pixmap.width > config.PDF_MAX_WIDTH:
+                scale = config.PDF_MAX_WIDTH / pixmap.width
+                pixmap = doc[i].get_pixmap(matrix=fitz.Matrix(dpi / 72 * scale, dpi / 72 * scale), alpha=False)
+
+            if ext in ("jpeg", "jpg"):
+                pixmap.save(str(img_path), jpg_quality=config.PDF_JPEG_QUALITY)  # JPEG 저장
+            else:
+                pixmap.save(str(img_path))
+            logger.debug(f"PDF 변환: {pdf_path.name} | 페이지 {i+1}")
+        img_paths.append(img_path)
+
+    doc.close()
+    logger.info(f"PDF 변환 완료: {pdf_path.name} | {len(img_paths)}페이지 | {dpi}dpi {ext}")
+    return img_paths
+
+
+def pdf_to_images_batch(pdf_paths: list[str],
+                        max_workers: int = config.OCR_MAX_WORKERS) -> dict[str, list[Path]]:
+    """
+    여러 PDF를 ThreadPoolExecutor로 병렬 변환
+    반환: {pdf_path: [img_path, ...]}
+    """
+    results = {}
+
+    def _convert(pdf_path: str):
+        return pdf_path, pdf_to_images(pdf_path)
+
+    with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = {executor.submit(_convert, p): p for p in pdf_paths}
+        for future in concurrent.futures.as_completed(futures):
+            try:
+                path, imgs = future.result()
+                results[path] = imgs
+            except Exception as e:
+                path = futures[future]
+                logger.error(f"PDF 변환 실패: {path} | {e}")
+                results[path] = []
+
+    return results