Abstract: A method for compressing a typographic font and recovering the original font from the compressed font, first extracts the outline of each glyph in the font in turn, and analyzes the outline to find all the dominant points. The curvature and slope(s) at the dominant points are collected for clustering at the font level. The dominant points are matched algorithmically and the outline is dissected into strokes. A normalized stroke descriptor is computed for each stroke so that similarities between strokes that are related by translation, rotation, reflection and scaling can be made explicit. Thus a stroke is described by its normalized descriptor along with translation, rotation, reflection and scaling parameters. After each glyph outline has been so analyzed, the curvature and slopes at dominant points are clustered. Also, the normalized stroke descriptors are classified so as to put similar strokes into the same class. Each of these classes is assigned a unique identifier.