Sketch Recognition Using Dynamic Bayesian Networks
Speaker: Metin Sezgin , Design Rationale group, CSAIL, MIT
As hardware such as Tablet PCs and PDAs that can capture and display freehand input become available to larger audiences, intelligent processing of the pen input gains more importance. The ability to recognize freehand drawings has the potential to revolutionize the way we interact with applications that currently use a traditional mouse/keyboard based interface for graphical input (e.g. CAD tools). Our goal is to realize this by building systems that can recognize freehand sketches (e.g. circuit diagrams, technical drawings). In this talk we show how online sketching can be seen as an incremental and dynamic process. We present a sketch recognition framework where we train dynamic Bayesian networks to build models of patterns present in users' stroke orderings and use these models to do efficient sketch recognition.