If this was my book I'd definitely remove the images because first-letter images are useless if you use text-to-speech to 'read' books.
The approach I'd take would be:
1. Hope that the creator had used a useful naming convention for each image, e.g. first-letter-A.jpg, first-letter-B.jpg, etc. Then I'd construct a regex to cycle through them, find/replace one at a time (not replace all).
2. If that wasn't possible I'd still remove the images, but manually, the hard way.
3. For your compromise option, css something like the following might be acceptable:
Code:
img.drop {height:1em; width:auto}