At some point the color gets flipped, we have to flip it again,
for mono.
Incidentally this exposed an unused
optimization. Should change the
first past to scan all images and
record monochrome xrefs, then optimize
JPEG and PNG, possibly adding
mono images to the monochrome
queue. Finally, do JBIG2 optimization.
It's unclear how exactly a
colormapped image gets to this
spot given the tendency of other
image processing tools to flatten
such images, but someone made it happen, so now we make sure
the image is okay.
Closes#262
Seems better to accept whatever leptonica rather than make detailed
assumptions about how it encodes the palette.
Experimented with setting FlateDecode on the palette but it seems to
expand it.
Turns out it does not transcode at all in this case, so probably going
to revert to transcoding PNG -> PNG. However if pngquant or similar is
done, this API will still be useful.
Leptonica does a better job of encoding them than Ghostscript, about -15%.
For a test file 450k worth of
PNGs was reduced to 388k with no loss of quality.
James R. Barlow