Find full set here: docs/00-letter-shapes.html (3) Frequently occurring kerning pairs I found this approach to generate way less noise than the popular vote before.Įxcerpt. Compared to the 11899 most "popular" kerning pairs, 11924 other possible ones could be added to the list. 20554 kerning pairs could be generated this way. More complex (serif) typefaces might need a higher resolution. For a grotesque sans serif typeface-as a proof of concept-a resolution of 3 by 5 units was sufficient for each side (-1, 0 or 1 for ascender, x-height, half-x-height, baseline, descender). Like everything that is considered here, this also depends on the design of the font. The values on the adjacent sides or two neighboring letters can be summed up to determine whether this is a possible kerning pair. The form of each letter can be described schematically by simple numerical values. See docs/01-kern-dump.html (2) Potential kerning pairsĪnother approach to determine possible kerning pairs is combinatorial. In Script 3G you can decide for yourself where you would like to make this cut, if desired. Which sounds a lot, but it leaves us with 11899 remaining most popular kerning pairs. In the end, I removed all values lower than 20 percent of the maximum value. It helped a lot to cut off the long tail at some place.
This use count contains a lot of noise caused by subjective decisions and/or group based kerning, possibly. This means, for the set of 300 characters considered, dazzling 80 % of the possible 90000 combinations are (more or less) used. 72377 unique kerning pairs were used in these fonts. It uses parts of the Adobe Type Tools, see directory "kernDump". But you can apply this script to any font collection you might want to look at yourself. As a sample I was content with the about 2700 font files of the Google Fonts Archive. Let’s start by looking at kern tables of existing fonts to collect more or less common kerning pairs statistically. the large sample size of texts in many languages. Some parts may also be useful for other works, e.g. In case you want to reproduce parts of this work: All scripts are self-explanatory and numbered according to the following article. Depending on your typeface’s style you don't have to kern each and everyone of them, of course. The first 1000 to 2000 pairs should have you covered sufficiently in all languages, see (4) for details.
If i understand the way it works, i will have no problem with OTF if i expand the kerning data. You're right, in fact in the TTF the total number of pairs is exactly 10920. #217 is the number of kerning pairs "grouped" (i don't know how to call the non-flat pairs).
If "flat" means the total number of kerning pairs Including the permutation with each member of a class, this is not the case.