Although we did extraction of glyph data from an image, this is not the last step in glyph recognition task. Applications dealing with augmented reality or robotics usually have a database of glyphs, where each glyph may have its own meaning. For example, in augmented reality each glyph is associated with a virtual object to be shown instead of a glyph, but in robotics applications each glyph may represent a command or direction for a robot. So the last step is to match the extracted glyph data with a database of glyphs and retrieve information related with the glyph - it's ID, name, whatever else.
To complete glyph matching step successfully, we need to keep in mind that glyphs can be rotated, so comparing extracted glyph data one to one with glyphs stored in database will not work. Finding a matching glyph in glyphs' database we need to do 4 compares of extracted glyph data with every glyph in the database - compare 4 possible rotations of extracted glyph data with the database.
Another important thing to mention is that all glyphs in database should be rotation variant in order to be unique regardless of rotation. If a glyph may look the same after rotation then it is a rotation invariant glyph. For rotation invariant glyphs we cannot determine their rotation angle, which is very important for applications like augmented reality. Also it may not be possible to find correct matching glyph in a database, if it contains few rotation invariant glyphs, which may look the same if one of them is rotated.
Below picture demonstrates some rotation variant and invariant glyphs. Glyphs (1) and (2) are rotation variant - if they are rotated they will look different always. Glyphs (3), (4) and (5) are rotation invariant - if rotated they will look the same, so it is not possible to detect their rotation angle. Also we may see that glyph (4) is actually same as glyph (5), but just rotated, so glyph database should not contain them both (even if it would be rotation variant glyph).