How to remove inline images from a pdf (to reduce the file size)

[Pmmoks]

I am able to extract most images with page.get_images(), and we can use the xrefs to delete the images (e.g. for the purpose of reducing the overall pdf file size).

We can also access inline images using page.get_text("dict")["blocks"] (if the block["type"] = 1). How would we also delete these images from the pdf data to reduce the file size?

[JorjMcKie]

Interesting question!
Inline images should be small as per recommendation in the PDF specs. They mention something like a 4K limit. But of course this is not mandatory.
PyMuPDF has not implemented (yet) direct access to individual inline images. They appear as an integral part of a page's appearance stream (the object(s) referenced under /Contents in the page's object definition) and thus have no way of identifying them individually (like with an xref).

Extracting images as you mentioned (part of the powerful text extraction method) works the same way for all supported document types - not only for PDF.

You can use a rather hacky approach to remove inline images, namely by "editing" the appearance stream. The following loops through a page's /Contents objects and removes everything between the start and the end of all inline images.

import pymupdf

doc = pymupdf.open("input.pdf")
for page in doc:
    page.clean_contents(sanitize=False)  # recommended to ensure a predictable stream formatting
    xref = page.get_contents()[0]  # there is only one now because of the clean
    contents = bytearray(doc.xref_stream(xref))  # modifiable version
    p0 = 0  # start of b"BI " - begin of inline image
    modified = False  # switch showing we encountered an image
    while True:
        p0 = contents.find(b"BI ", p0)  # look for image start
        if p0 < 0:
            break
        p1 = contents.find(b" EI", p0)  # look for end of image
        if p1 < 0:  # should not happen!
            print("stream format error - stopping")
            break
        contents[p0 + 3, p1] = b""  # remove image content
        modified = True
        p0 += 3  # step behind "BI " marker

    if modified:
        doc.update_stream(xref, contents)

I hope it works! Didn't test it.

[JorjMcKie]

You can also try to use redactions for image removal:

import pymupdf
doc = pymupdf.open("input.pdf")
for page in doc:
    page.add_redact_annot(page.rect)
    page.apply_redactions(

        # remove images    
        images=pymupdf.PDF_REDACT_IMAGE_REMOVE,
            
        # don't touch graphics
        graphics=pymupdf.PDF_REDACT_LINE_ART_NONE,

        # don't touch text
        text=pymupdf.PDF_REDACT_TEXT_NONE,                 
    )
doc.ez_save("images_stripped.pdf")

I think this removes all images - whether or not inline.
Please note that annotations may also contain images: they are not removed by either of these approaches, because by definition, annotations do not count as page content.
To also remove annotation images, there are these options:

1. Execute doc.bake(). This will convert annotations (and fields) to standard page content. Then the above methods will work.
2. Iterate through annotations and remove any that contain an image (not shown here).