James Arvo and Kevin Novins
We present a new approach to managing the appearance of handwritten symbolic information for panbased systems in which text is recognized and gradually transformed into clean typography. The approach is best suited to pen-based symbolic input and manipulation in which the user interacts directly with the display. Once a collection of user-drawn strokes has been recognized as comprising a specific symbol, the strokes begin a gradual metamorphosis into the corresponding raster image from an appropriate font. The transformation retains legibility at all intermediate stages and can proceed smoothly at widely differing rates. The metamorphosis is accomplished by an energy-based stroke correspondence algorithm, followed by level-set shape transformation that gradually introduces characteristic features of a given font, such as serifs and spurs. This technique leads to a qualitatively different form of pen-based interaction in which the user enjoys the benefits of character recognition and highly legible typefaces without abrupt changes in appearance or positioning of the text. The interplay between hand-written symbols and pre-defined raster fonts suggests a number of new methods for correcting character recognition errors; for instance, the recognizer can be given additional hints by "touching up" an incorrectly recognized symbol.