pytesseract

pytesseract is a thin Python wrapper around the Tesseract OCR command-line engine, letting Python code extract text from images without shelling out manually. It supports all Tesseract image formats handled by Pillow/Leptonica (JPEG, PNG, GIF, BMP, TIFF, and more) and returns plain text, structured bounding-box data, TSV, or searchable PDF/HOCR/ALTO output.

Because it’s a wrapper rather than a reimplementation, output quality and language support are entirely determined by the underlying Tesseract installation, which must be installed separately on the host system. This makes pytesseract lightweight, but it also means Tesseract itself (and any trained language data) is a required system dependency.

What You Get

• image_to_string() for plain-text OCR output from an image
• image_to_boxes() and image_to_data() for character/word bounding boxes and structured TSV output
• image_to_pdf_or_hocr() for generating a searchable PDF or HOCR/ALTO document from an image
• Support for all image formats Pillow/Leptonica can read (JPEG, PNG, GIF, BMP, TIFF)
• Configurable Tesseract flags (language, page segmentation mode, OCR engine mode) passed straight through

Common Use Cases

• Extracting text from scanned documents or receipts in a Python data pipeline
• Building searchable archives from scanned PDFs by generating a text layer with image_to_pdf_or_hocr()
• Automated form/invoice data extraction combined with image preprocessing (OpenCV, Pillow)
• Quick OCR scripting/prototyping where a full computer-vision pipeline is unnecessary

Under The Hood

Architecture: the entire wrapper lives in pytesseract/pytesseract.py (~650 lines) — it writes the input image to a temporary file, shells out to the tesseract binary via subprocess, and parses stdout/temp-file output back into Python strings, dicts, or bytes depending on the called function; run_and_get_output() centralizes the subprocess invocation used by all the image_to_* helpers. Tech Stack: pure Python with no compiled extensions, depending on Pillow for image handling; the actual OCR work is delegated entirely to the external Tesseract executable, which is a required system-level dependency, not a Python package. Code Quality: tests/pytesseract_test.py covers the public API against fixture images, and CI runs via GitHub Actions with pre-commit hooks enforcing style — reasonable coverage for what is fundamentally a subprocess wrapper. API Design: the API is a small set of image_to_* functions with a consistent first-argument (image) and keyword-based Tesseract config passthrough, making it easy to get a first OCR result in one line, though power users need to know Tesseract’s own CLI flags to unlock advanced behavior.