In OO/pdfimport you can do it by clicking the background image element and delete it with the del key. Cumbersome if you have lots of images, but at least you actually see what you are doing and what you're getting... and pdfimport might even be able to let you manually fix a typo here and there?
In vim you find the image element (it's the largest one with 500kb or whatever), note the line number where it starts, search for endstream, note the line number where it ends, and then you use the command :line1,line2d to delete the whole area. Then you save and the image is gone.
I could write a script that does the vim step automatically - if I find the time, and if you use Linux/Python cmdline.
I don't know what to edit to make the text visible though. You'd have to replace the font somehow but how? Might have a go at willus PDF later to see what he changed :P