| """Classical OCR page reader: a `textsource.PageReader` that transcribes rendered pages on CPU with |
| Tesseract and makes no model call. |
| |
| Shared by the two CPU backends: #84 pairs it with rules-only extraction, #85 with a local LLM extractor, |
| so it is defined once here and reused rather than duplicated. The default engine runs Tesseract through |
| `pytesseract`, which needs the `tesseract` binary installed on the host. The engine is injectable |
| (`ocr_page`), so a test passes a fake and a deployment could swap in another classical engine (PaddleOCR) |
| without touching the reader. |
| """ |
|
|
| from __future__ import annotations |
|
|
| from pathlib import Path |
| from typing import Callable, Optional |
|
|
| |
| |
| PAGE_SEPARATOR = "\n\n" |
|
|
|
|
| def tesseract_ocr(path: Path) -> str: |
| """Transcribe one page image with Tesseract. Imports `pytesseract` and Pillow lazily, so importing |
| this module needs neither the packages nor the tesseract binary until a page is actually read.""" |
| import pytesseract |
| from PIL import Image |
|
|
| with Image.open(path) as image: |
| return pytesseract.image_to_string(image) |
|
|
|
|
| class ClassicalPageReader: |
| """A `textsource.PageReader` that OCRs a report's rendered pages on CPU and joins them into one report |
| text, with no model call. The CPU backends read pages through this in place of the vision-model |
| reader. `ocr_page` transcribes one page image and defaults to Tesseract; a test injects a fake.""" |
|
|
| def __init__( |
| self, |
| *, |
| ocr_page: Optional[Callable[[Path], str]] = None, |
| page_separator: str = PAGE_SEPARATOR, |
| ) -> None: |
| self._ocr_page = ocr_page or tesseract_ocr |
| self._page_separator = page_separator |
|
|
| def read(self, page_images: list[Path]) -> str: |
| parts: list[str] = [] |
| for path in page_images: |
| text = self._ocr_page(Path(path)).strip() |
| if text: |
| parts.append(text) |
| return self._page_separator.join(parts) |
|
|
|
|
| def tesseract_word_boxes(path: Path) -> tuple[list[dict], int, int]: |
| """The words Tesseract reads on a page, each with its pixel box, plus the image size. Lazy imports, |
| so importing this module needs neither pytesseract nor the tesseract binary until a page is read.""" |
| import pytesseract |
| from PIL import Image |
| from pytesseract import Output |
|
|
| with Image.open(path) as image: |
| width, height = image.size |
| data = pytesseract.image_to_data(image, output_type=Output.DICT) |
| words: list[dict] = [] |
| for i, text in enumerate(data["text"]): |
| if text and text.strip(): |
| words.append( |
| { |
| "text": text, |
| "left": data["left"][i], |
| "top": data["top"][i], |
| "width": data["width"][i], |
| "height": data["height"][i], |
| } |
| ) |
| return words, width, height |
|
|
