A PDF page is a drawing program: it can combine text, vector paths, images, clipping, transparency, annotations and transformations. Rendering executes that composition into one visible pixel rectangle. Extracting embedded images instead copies or decodes individual image objects and does not automatically reproduce the page around them. The two jobs have different outputs and should not share one promise.
Render when you need the visible page
Use page rasterization for a website preview, presentation thumbnail or system that accepts an image of the whole page. PageRaster follows the crop box and rotation reported by PDF.js, paints sequentially, and names each result by source page number. Links, form interactivity, scripts, attachments, layers, audio and selectable text are not preserved in the image. Normal PDF.js painting may include visible annotation or form appearances, so inspect them in the output.
Extract when you need an original asset
If the goal is the original photograph stored inside a PDF, use a dedicated image-object extractor that can report masks, colour spaces, transformations and repeated object references. PageRaster deliberately does not expose that workflow. It also does not OCR text, rebuild a document, remove a background or recover a pre-compression source. A rendered crop of a photograph includes the page renderer’s scale and composition, and may differ from the embedded object’s native pixels.
Verify the downloaded pixels
Open every downloaded PNG or JPEG in the application that will use it, confirm the page number and measured width and height, and compare visible text, fonts, colours, transparency and annotations with the authoritative PDF. Keep the PDF as the source. Raster output is a browser-rendered snapshot, not proof of vector, print, archival, accessibility, legal or semantic equivalence.