Drumroll....

[BlackStar]

I believe a screenshot of this page for a few different fonts would be a good start: http://www.infinality.net/files/font.html

I'd suggest pictures of fonts like: "Arial", "Times"/"Times New Roman", "Lucida Grande", "Menlo", "Monaco", "Verdana", "Trebuchet MS", "Candara", "Calibri". (Some of those will only be available if you happen to have MS Office installed).

[cartman]

Can't find Candara and Calibri, 12-13px version is @ http://cekirdek.pardus.org.tr/~ismail/m ... 2-13px.zip

[BlackStar]

Here are comparisons for Arial and Lucida Grande between Infinality and Mac:

Arial (Infinality)


Arial (Mac)


Arial (Linux, unhinted)



Lucida Grande (Infinality)


Lucida Grande (Mac)


Lucida Grande (Linux, unhinted)


The Mac shots are slightly blurry (check capital E and F for Arial and letters { and 1, for instance).

Infinality generally looks cleaner but kerning although kerning appears a little tight (Lucida fE touch due to the aggressive hinting - f is supposed to be slightly higher, as evidenced in the Mac image?) I also don't like how the bottom parts of the small 'a' letters look slightly smudged.

Edit: added unhinted Linux screenshots. I must admit they look much better than I expected on my 140dpi netbook monitor (they are still horrendous on a 23'' screen, though).

[markser]

@Infinality: You said you have a Windows 7 installation -- just download Safari for Windows, it comes with Mac-style font display (slightly configurable in the options).

[infinality]

Thanks cartman!

Ok, here is a screenshot of Arial using the following env vars along with the "Apple" style rendering in local.conf (on the left). On the right is cartman's screenshot. If you zoom in, you can see that the LCD filtering is a bit darker on the actual apple shot. In fact, it looks like it may even be a filter done after the LCD filtering. My question is this: Does the slight difference between them justify adding another filter, considering they look very close as is?

Code: Select all

INFINALITY_FT_AUTOFIT_ADJUST_HEIGHTS=false
INFINALITY_FT_EMBOLDEN_MAINTAIN_WIDTH=true
INFINALITY_FT_AUTOFIT_FORCE_SLIGHT_HINTING=true
INFINALITY_FT_AUTOFIT_STEM_SNAP_LIGHT=true
INFINALITY_FT_SHARPEN_ITALICS=false
INFINALITY_FT_AUTOFIT_EMBOLDEN_LIGHT=true
INFINALITY_FT_FILTER_PARAMS=05 25 40 25 5
INFINALITY_FT_WINDOWS_STYLE_SHARPENING_STRENGTH=0
INFINALITY_FT_CHROMEOS_STYLE_SHARPENING_STRENGTH=0
INFINALITY_FT_STEM_ALIGNMENT_STRENGTH=0
INFINALITY_FT_AUTO_AUTOHINT=true
INFINALITY_FT_AUTOHINT_STEM_SNAPPING_STRENGTH=0
INFINALITY_FT_STEM_FITTING_STRENGTH=0
INFINALITY_FT_FRINGE_FILTER=false
INFINALITY_FT_ENHANCED_EMBOLDEN=true
INFINALITY_FT_PSEUDO_GAMMA=20 80


[BlackStar]

(I've edited my previous post to add unhinted Linux screenshots).

@infinality: there's something more going on here than meets the eye. Either Mac hints slightly in the y-direction or there is a dpi difference.

Compare the 12px small 'a' letters. The Mac one is is slightly *taller* (a few percent) so that the top stroke falls on a whole pixel. On tthe Infinality screenshot, the top stroke falls just below a whole pixel, causing it to be mostly gray (due to anti-aliasing).

The same difference can be observed in the middle-left of the 'a' letter: the curve on the Mac is slightly higher than on Infinality, so that (my guess) it falls on a whole pixel.

What I *believe* is going on: Mac doesn't use BCI hinting but attempts to deform glyphs slightly to better fit the pixel grid. If I had to guess, this is a form of y-direction autohinting with changes capped to at most 25% of a pixel (something like what the Infinality autohinter does). I believe most of the differences can be accounted by this change only (glyphs look slightly darker because more pixels are aligned to the grid - and are hence black rather than anti-aliased), without a secondary filter.

What do you think?

Edit 2: The 12px "fE" combination reveals something *very* interesting. Infinality anti-aliasing looks much stronger than Mac.

Three possible explanations:
- a different FIR filter is in use
- a secondary sharpening pass is used in the Mac
- the Mac rendering is oh-so-slightly hinted so that the top part of the letters is closer to a whole pixel (hence much lighter anti-aliasing is taking place).

Edit 3: another possible explanation: Mac hints but does *not* alter stem thickness. What this means is that the top part of fE is indeed snapped to a pixel but the stems are thicker than 1px (but less than 2px) so some anti-aliasing is taking place. Freetype, on the other hand, would reduce the stem to just 1px for perfectly clear (but less faithful) rendering.

I don't know if this is true, but it *could* explain some of the differences!

[infinality]

Yes, I noticed the same thing with the strokes falling in different spots. It's quite possible they are doing a very simple form of hinting, at most 25% of a pixel, which explains why there are still grey fringes on some horizontal stems. (What I don't get though is if they are going to "kind of" align horizontal stems, why not just "actually" align them?) I can experiment to see if I can determine exactly what they are doing.

I do think that they are either implementing FIR filtering differently, or running a filter afterwards to darken dark parts and lighten light parts. I'll try a sigmoidal filter post-FIR and see if that makes it closer to the actual apple shot.

[BlackStar]

Yes, I noticed the same thing with the strokes falling in different spots. It's quite possible they are doing a very simple form of hinting, at most 25% of a pixel, which explains why there are still grey fringes on some horizontal stems. (What I don't get though is if they are going to "kind of" align horizontal stems, why not just "actually" align them?) I can experiment to see if I can determine exactly what they are doing.

Hm, what if they do actually align the stem, but it just happens that it doesn't fit within a single pixel (i.e. a parts of it 'spill out' to the neighboring pixels)?

I do think that they are either implementing FIR filtering differently, or running a filter afterwards to darken dark parts and lighten light parts. I'll try a sigmoidal filter post-FIR and see if that makes it closer to the actual apple shot.

Hm, I'm looking at the images zoomed-in and the Mac ones look significantly heavier. IIRC, you have implemented the infrastructure for a 2nd filtering pass? Would be interesting to see the results.

Edit: how are you testing your builds? Do you simply build freetype and preload it to some application or have you set up some crazy chroot system or something?

[cartman]

At 12px Mac results are easier to read, looks more crisp but at 13px for example its hard to differentiate.

Thanks!

[cartman]

[quote="infinality"]

Code: Select all

INFINALITY_FT_WINDOWS_STYLE_SHARPENING_STRENGTH=0
INFINALITY_FT_STEM_FITTING_STRENGTH=0


These options doesn't exist in current infinality code, please release new stuff!