OK ... I think
pdflatex is the missing piece here.
The OP said he has looked into
pdftk. Let me add to that
pdfimages. These, together with
pdflatex are the pieces of a solution.
pdfimages -f 4 -l 20 -j -png target.pdf imageroot
In the example code above,
pdfimages looks through pages 4 through 20 of
target.pdf and extracts all images into files with names beginning
pdftotext. I recommend the
-layout option which does a great job keeping the document human readable.
pdftotext -layout $1.pdf $1.txt
The OP's objection to the
imagemagick solution offered by pidosaurus is that
an image does not have extractable text. With the utilities I outlined, the OP will now have all the images as well as all the extracted text, and page numbers and contents are retained by the
-layout option. The OP could identify the correct page of text and chuck it into a
.tex file which ends with an
%includegraphics directive and refers to the replacement picture by filename. You then
pdflatex this and end up with a new single-page .pdf to insert into the rest of your document with
pdftk. If you knew where in the text of the original page the image resided, you can
%includegraphics [h] and get the image in exactly the right place.