Incorporating Context and User Feedback in Pen-Based Interfaces

Contents

We propose a joint probabilistic model for grouping and labeling hand-drawn ink strokes. We demonstrate that simultaneous grouping and labeling yields superior accuracy to labeling alone. Our probabilistic formulation has many advantages, exact inference is feasible, and we obtain confidence estimates. We show how to incorporate user feedback by conditioning our model, and discuss different types of inference tasks suited for various user interactions.