Quote:
Originally Posted by twynn92
Also make sure the file extensions match too, but hopefully that will never be an issue since Amazon converts already anyways.
|
Aaand it's a good thing I did perform the check, as the third book I did had some GIFs in the KF8 which were converted to JPGs. I opted to keep the original GIF images, as (1) that doesn't require me to go into the text and mass replace .gif to .jpeg and (2) GIFs are lossless while JPGs are lossy. I'm not sure if that makes a difference since I'm sure Kindle-published books go through a conversion pipeline anyways. @jhowell Do you know anything about the GIF/JPG discrepancy between KF8/KFX?
The other thing to look out for is the cover image discrepancies. In other words, sometimes there are two references to two different images in the KF8, one being an IMAGE element within an SVG, and sometimes there are two references to the same image in the KFX, also duplicating the pattern of SVG/IMAGE tags and regular IMG tag. I think it's safe to (1) replace the cover image in cover_page.xhtml and (2) duplicate the same image in part0000.xhtml (if there is one) as there's precedent for that in the KFX source. Note that sometimes part0000.xhtml doesn't have a cover image and just starts with the frontmatter of the book, so keep an eye out to make sure there there are no unintended offsets.
Quote:
Originally Posted by jhowell
If I was going to do this I would add a step to check the accuracy of image matching between formats by finding an algorithm to determine image similarity and using it to automatically verify that the candidate substitutions are close matches.
|
I didn't know there was such a thing, else I probably wouldn't have done so many manual checks via the single HTML text comparison as outputted by Pandoc. I mean aside from the cover image discrepancies, and the GIF vs. JPG difference, everything else is a one-to-one match, though the KFX code has too many necessary classes...
Thanks for giving me the opportunity to bounce ideas off of you. As you can tell, I'm no programmer, but do know bits and pieces to do just enough and get the intended output. This was more an exercise of curiosity than anything else, but good to know it's definitely possible with only one or two caveats.