Blue Zones in FreeType 2

Introduction & Background:

It is common in Latin scripts to put the top of rounded glyphs, like "O", "C" or "Q", above the top of flat glyphs like "T" or "Z" by a small distance. This variation is an aesthetic choice, used to improve the appearance of text. Without it, rounded glyphs often appear a tiny bit smaller to the human eye.

However, it is very important, when grid-fitting the extrema of glyphs, that the top of rounded and flat glyphs be snapped to the same pixel grid. Otherwise this small distance might be stregthen resulting in very unpleasant artefacts, depending on the current pixel size, as in the following picture:

image of un-aligned tops and bottoms

To address this problem, the zone of points whose y coordinate is located between the top of flat glyphs (called the zone's reference coordinate) and the top of rounded glyphs (called the zone's overshoot coordinate) is treated in a special way. Such zones are called "blue" zones. Usually, the following zones are noticed and processed in Latin scripts:

  • the top of capital letters
  • the top of small letters
  • the bottom of capital letters
  • the bottom of small letters
  • the bottom of "legged" small letters like "p", "q", "g", "j" & "y"

Note that some other blue zones might be defined to process specifically certain symbols like:

  • enclosure symbols like "(", ")", "[", "]", "{", "}" which typically have more or less the same height in a digital font
  • some special math symbols, like the integration symbol, the sigma symbol, and others. Figures may also be important for math uses.

Note that when the overshoot coordinate is above the reference coordinate, the blue zone is said to be a "top zone". This is the case for the top of capital letters, small letters, enclosures, etc.. Similarly, when the overshoot is under the reference, the blue zone is said to be a "bottom zone".

The FreeType Auto-Hinter computes the blue zones automatically. The purpose of this page is to present the technique used to do that, as well as the one used later to grid fit points within this blue zone. Note that for now, we only consider Latin scripts. Other scripts that may require blue zone alignement might be added in the future (e.g. arabic, hebrew, etc..)

Automatic Computation Algorithm:

It is assumed that the font is well designed. This means, in the case of the "capitals top" blue zone that all flat characters (like "Z", "T", "H" etc..) have the same height (given a certain threshold); moreover, that all rounded characters (like "O", "C", "S", "Q", etc..) have the same height.

Because of differences in design between fonts, we cannot assume however that a well-know character like "Z" has always a flat top, or that a "C" has two rounded vertical extrema.

For a given blue zone, the technique used simply loads a set of well-known characters that present some "features" related to the zone. For example, the characters of "HEZLOCUS" are loaded to compute the "capitals bottom" zone.

For each zone, the algorithm uses two lists of coordinates that are initially empty. Each character in this "zone set" is then analyzed for either its upper or lower extrema, depending on the zone's orientation. If the extrema is rounded, its coordinate is added to the first list, if it is flat, it is added to the second list.

When all characters in the set are loaded and analyzed, the median of each list is taken to compute the reference and overshoot position of the blue zone. In the case where one of the list is empty, the two positions are both set to the median of the remaining list.

By keeping a small sets of well-chosen characters for each zone, we avoid loading each glyph in the font for analysis, speeding considerably the computation compared to a real global analysis (normally performed by building the histograms of flat and rounded tops of each glyphs, and detecting "peaks"; this detection phase can be tricky if the font contains many symbols that are not part of a Latin script..).

Blue Zone Grid-Fitting:

For each pixel size, the blue zones need to be grid-fitted to the current scale. This is done by normally grid-fitting the reference position of each zone, the positioning the overshoot position relative to this reference grid-fitted coordinate.

As usual with blue zone in digital typography, if the size in pixel space of the blue zone falls under a certain threshold (like ONE pixel), the blue zone is "snapped", which means that the overshoot is grid-fitted to the reference position itself (along with each horizontal edges between them).

Blue Edges detection:

We need to "snap" some of the horizontal edges detected in the feature-detection pass to the grid-fitted reference and overshoot position of the blue zones.

We select the edges according to the following rules. Note that the TrueType filling convention is used in this example (directions must be inversed for Type 1 fonts):

  • Only right-wards edges can be snapped to top zones, and left-wards edges can be snapped to bottom zones. This in order to avoid snapping the inner contours of glyphs to a blue zone. (invert left and right for Type 1 fonts).

  • The edge can be snapped either to the zone's reference or overshoot coordinate depending on its position relative to the zone and its roundness flag. For example, for top zones:

    • A round edge will be snapped to the pixel overshoot if it is within a certain threshold of the overshoot in design space.

    • A flat edge will be snapped to the pixel reference if it is within a certain threshold of the reference in design space.

    • To avoid weird artefacts, a round edge may NOT be snapped to a blue position when it is under the reference point.

    • Similarly, a flat edge may NOT be snapped to a blue position when it is over the overshoot.