Re: [ft-devel] Splitting up the GSoC project (Re: Freetype-devel Digest, Vol 148, Issue 36)

[Hin-Tak Leung]

--------------------------------------------
On Mon, 29/5/17, Arvinder Bhathal <address@hidden> wrote:
 
> > 2.
 pick any one font, actually generate 26 images of A to Z,
 for two versions of freetype, to get a feeling of what sort
 of bitmap differences you are likely to get.
 
> Although I'm unsure of the
 feasibility of it, my initial thought was
 to
 sum the pixel intensity differences between the baseline and
 test
 glyphs. Supposedly, these values could
 be stored for every glyph
 (although maybe
 not for 12 million) in an index, and we could consider
 these values when displaying the glyphs in the
 browser. This could
 help focus on the pairs
 with the greatest differences. However, as
 you've said in your second suggestion, a
 demo test will help determine
 what sort of
 comparison method will do.
 

I suspect for most glyphs in most fonts, you would get identical bitmaps for 
B/W and 8-bit rendering between not-too-distant freetype versions. (LCD is a 
different matter lately). So the actual numbers of glyphs you need to examine 
at all, is a small fraction of of a large number. Hence you might need to test 
quite a few fonts before you get "interesting" differences.

As for quantifying differences - just summing pixel difference isn't a good 
measure. The differences is likely to be corner pixels or whole edges; i.e. a 
corner or a whole edge has moved slightly. I'd probably suggest somehow taking 
connected-ness into account - e.g. if you draw a square of 4x4 pixels, the 4 
corner pixels being slightly darker/lighter, is less important than a whole 
edge of 4 pixels being darker/lighter. So you weight the difference by how many 
nearby pixels also show differences. 

A complicated shape will have a lot of sharp corners and the many corners may 
go darker/lighter, but these are less important than one stem of a simple shape 
going thicker. If you only count number of pixels differing, the first case 
might have a higher "score" of differences.

Anyway, I think it is best your people have a go at looking at say, differences 
of bitmaps of say A-Z of one font you choose, and see what sort of difference 
you can get, to get a first-hand feeling about it.