Extracting Images from a PDF

Where we left off

Every image so far arrived as its own standalone .png file. Real documents rarely cooperate: a scanned notebook page, a slide deck exported to PDF, a report with inline figures, all deliver images embedded inside a PDF page, not as separate files sitting in a folder. This lesson handles that case using circuit-board-notebook.pdf, a one-page PDF fixture whose single page has one embedded image (see fixtures/README.md), extracting that image before handing it to caption_image() exactly as if it had been a standalone file all along.

The code, piece by piece

from pypdf import PdfReader

reader = PdfReader(PDF_PATH)
page = reader.pages[0]
for image_file in page.images:
    image_bytes = image_file.data

pypdf (already installed transitively in this project, via markitdown/docling's own dependencies) exposes each page's embedded images through page.images, an iterable of ImageFile objects. .data is the raw image bytes, exactly the same shape caption_image() has expected since Lesson 4, whether those bytes originally came from reading a .png file directly or from unpacking them out of a PDF page changes nothing downstream.

caption = caption_image_from_bytes(image_bytes, mime_type="image/jpeg")

One small change from caption_image(): the function now takes raw bytes and a MIME type directly instead of a Path to read, since there's no standalone file to read from, only bytes already extracted from the PDF. page.images re-encodes embedded images as JPEG internally (visible in image_file.name ending in .jpg), so the MIME type passed to Part.from_bytes needs to say image/jpeg, not image/png, matching what's actually in the bytes.