In this thesis, a pen-based interactive table and diagram input system, named PenDraw, is developed. The system consists of two major subsystems, table input system and diagram input system. The table input system constructs regulated table form from roughly sketched rectilinear lines and the diagram input system builds regulated diagrams with connecting lines from roughly hand-drawn sketches.
For table input, the system straightens hand-sketched lines and makes intelligent guesses about where the user's intended end points should be located. If a given pattern is recognized as a table object, operations specific for table object such as expanding, reshaping and aligning table can be performed using gesture and pen dragging.
The diagram input system converts hand drawings to lines and then estimate whether the lines may form a loop-structure. If they do, the system classify it into one of 17 internal loop models and displays it to corresponding regulated form. Internal models can be trainned by a few examples. While most existing systems impose some kinds of restrictions on drawing style in terms of number of strokes and/or order of strokes, PenDraw is intended to impose no restrictions on drawing style.
To evaluate PenDraw's diagram recognition accuracy, an experiment is conducted with 10 sets of hand-sketched diagrams of total 744 symbols. The classifier was trained with only 10 examples, but it yields 93% of recognition rate. It has been demonstrated that such a pen-centric interaction style seems natural and effective for making tables and diagrams.
