Previously we attempted to reinsert all PNGs, but it appears to be
unlikely that Leptonica's API is actually capable of optimizing the PNG
before it inserts it.
In any event qpdf has gained image optimization capabilities as well
which we coudld borrow.
There seems to be version to version inconsistencies between
Leptonica's photometric interpretation of 1bpp images, in
particular commit a0692307 introduces a change to force transcoding
in this situation.
However, I never entirely got to the bottom of where the problem
is, and in any event 1bpp images are probably better optimized
by JBIG2 than pngquant, so we're going to stop running them through
pngquant.
During some past refactor it appears we now end up treating
all JBIG2 images as if they appeared on the last page in the
file. This bug had no visual side ffects but probably led to
suboptimal JBIG2 encoding.
If the color count of an image is low such as when black and white
documents are scanned in color, PNG with lossy quantization may
produce a superior encoding to JPEG. This is expensive to test however.
Previously we opened pngquant-compressed PNGs with transcoding
because the transcode free function in Leptonica didn't seem to
work. This mean Leptonica may have thrown away the hard of
pngquant if didn't understand the encoding.
This change resolves the issue and allows us to open PNG encoded
data and insert it into a PDF without transcoding. Should improve
encoding quality.
* Fix unreachable code
This fixes an issue reported by LGTM.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
* Remove unused imports
This fixes several recommendations from LGTM.
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Soft masks are only allowed to be of colorspace DeviceGray so we
shouldn't use pngquant on them. For now, avoid this exceptional
case by excluded soft masks from optimization.
James R. Barlow
James R. Barlow
mawi
Stefan Weilandjbarlow83