Yeah, for the images incorrectly marked as covers, if I open the epub in sigil I see for example for img1.jpg (which is NOT a cover) Add Semantics -> Cover Image is checked (sometimes also for one of the html files Semantics -> Cover checked). I'm not sure why was the file done this way but I suppose this I will have to edit each of them.
On the other hand, I've imported the same small collection of epubs (146 files) in Sony Reader for PC and none of the incorrect covers appeared as cover. (Possibly, the Sony software accepts a valid cover only if the file name is cover.jpg).
Too bad I don't know how to copy the files from the Sony software (I have no ideea where are the books stored and IF they are accessible).
I've tried the suggestion not to read the metadata from file and while indeed no cover was generated this way, the import is a mess, with titles and authors revers for a lots of books (they do not have consistent names).
All this trouble only to be able to organize 146 books in folders with correct author names. I suppose I could have done it by hand easier ... lol.
I guess I will have to give up searching a way NOT to generate covers from first page.