His final solution (500 equidistant points per glyph) is very clever though visually, it results in putting unnecessary emphasis in areas of variability. For ex

His final solution (500 equidistant points per glyph) is very clever though visually, it results in putting unnecessary emphasis in areas of variability. For example, look at the bottom of uppercase 'Q' or the middle of lowercase 'f' - notice how it appears to be a thick knot instead of being the same width as the rest of the letter. This is most certainly because averaging the perimeter points ends up enlarging the enclosed area. The more darts you throw at a dartboard, the larger the area you will cover. While he mentions that he solved the issue of excess variability (lowercase 'g') by selecting a subset of similarly styled glyphs, I don't think he did anything to shrink the knots. Off the top of my head, one thing that could help solve the knot problem is treat the serifs and sans-serifs as two different subsets and make two Average typefaces. Additionally, he could calculate 'average distance to X nearest points in reverse direction'. By that I mean, a point on a curve pointing down should look at all the points that point up (meaning they are on the other side of the filled polygon or hole), and find the 5-10 closest points and take the average. This could tell you how thick the line or hole is at that point. Then use this as the weight to come up with a weighted average of all the other parameters for each point. All-in-all, this is a fantastic project and the weeks and months of independent research is time well spent. Hacking for fun is wonderful.