GRAPHICAL INPUT THROUGH MACHINE RECOGNITION OF SKETCHES
Christopher F. Herot
Architecture Machine Group, Department of Architecture
Massachusetts Institute of Technology, Cambridge, Massachusetts
A family of programs has been developed to allow graphical input through continuous digitizing. Drawing data, sampled at a high and constant
rate, is compressed and mapped into lines and splines, in two and three dimensions. This is achieved by inferring a particular user's intentions
from measures of speed and pressure.
Recent experiments have shown that even the most basic inference making cannot rely solely upon knowledge of the user's drawing style, but
needs additional knowledge of the subject being drawn, the protocols of its domain, and the stage of development of the user's design. This
requirement implies a higher level of machine intelligence than currently exists. An alternate approach is to increase the user's involvement in
the recognition process.
Contrary to previous efforts to move from sketch to mechanical drawing without human intervention, this paper reports on an interactive
system for graphical input in which the user overtly partakes in training the machine and massaging the data at all levels of interpretation. The
initial routines for data compression employ parallel functions for extracting such features as bentness, straightness, and endness. These
planned for implementation in microprocessors.
Results offer a system for rapid (and enjoyable) graphical input with real-time interpretation, the beginnings of an intelligent tablet.
There are many areas of human endeavor which could benefit from the use
of computer aids if there existed an effective means of communicating
about these tasks with a machine. While the field of computer graphics arose
to fill that need, it has too often added a new level of complexity. In
computer-aided design, for instance, the process of "digitizing" is suffi-
ciently cumbersome to delay its application until a relatively complete
design has been produced by the human designer. The result is usually more
akin to computer-aided evaluation or manipulation than to computer-aided
design. The research described here is motivated by the desire to involve the
computer in the early stages of the design process, where the feedback
generated by the machine can be most useful. The medium chosen is free-
hand sketching, as done with pencil and paper, as could be done at a data
tablet. A machine is postulated to be looking on while the user is sketching.
It could make inferences not only about the meaning of the sketch but also
about the user's attitudes toward, and uncertainties about, his design.
This approach offers its own unique set of problems and solutions, since the
data available to the machine are at once plentiful and incomplete. While
the ultimate implementation assumes a near-human intelligence on the part
of the machine, far off in time, there are many interesting things to be
learned along the way.
Our previous experiments[1, 2, 3, 4] have been directed toward the