Bug 697351

Summary:	/Subsample filter needs to make non-integer scales integer for monochrome images
Product:	Ghostscript	Reporter:	yiwei.she
Component:	PDF Writer	Assignee:	Ken Sharp <ken.sharp>
Status:	RESOLVED FIXED
Severity:	normal
Priority:	P4
Version:	9.20
Hardware:	Macintosh
OS:	MacOS X
Customer:		Word Size:	---
Attachments:	monochrome subsample bug test file

Description yiwei.she 2016-11-16 09:02:40 UTC

Using the command:
gs \
  -o output.pdf \
  -sDEVICE=pdfwrite \
  -dDownsampleMonoImages=true \
  -dMonoImageDownsampleType=/Bicubic \
  -dMonoImageResolution=200 \
   input.pdf

Gives the following error (on every page):

The only Downsample filter for monochrome images is Subsample, ignoring request. Subsample filter does not support non-integer downsample factor (3.017893) Failed to initialise downsample filter, downsampling aborted. 

According to current documentation at:
http://www.ghostscript.com/doc/9.20/VectorDevices.htm#PDFWRITE,

the Bicubic filter should be supported for monochrome.

Comment 1 yiwei.she 2016-11-16 09:03:24 UTC

Created attachment 13149 [details]
monochrome subsample bug test file

Comment 2 Ken Sharp 2016-11-16 09:11:56 UTC

(In reply to yiwei.she from comment #0)
> According to current documentation at:
> http://www.ghostscript.com/doc/9.20/VectorDevices.htm#PDFWRITE,
> 
> the Bicubic filter should be supported for monochrome.

No, that's not an option, the documentation clearly has a typo which I'll also fix.

Bicubic filtering requires intermediate levels between black and white, so the image would no longer be monochrome, and for imagemasks would no longer be an imagemask. Since the point of downsampling an image is to reduce its size, converting it from 1 bit per pixel to 8 bits per pixel isn't really an excellent plan.

Subsampleing *is* the only option which retains the monochrome nature of monochrome images, but I agree that in this case we should either permit a degree of freedom or, since its one level for all images no matter what the size of the image or its original resolution, clamp to the nearest integer value.

Comment 3 yiwei.she 2016-11-16 11:56:13 UTC

The subsample filter also not rounding when the ration is within .1 of an integer; in the error the ratio is (3.017893).

Comment 4 Ken Sharp 2016-12-02 11:07:23 UTC

commit b880332b899e0e59d17c7e48033e5cc816e5a831 will now force the downsample factor to an integer when the input is a 1 BPP image or imagemask (so we cannot shift to a Bicubic filter) and the downscale factor is not precisely an integer.

Also fixed the mistake in the documenttion.