Exploring Better Techniques for Diagram Recognition

A critical component of diagramming sketch tools is their ability to reliably recognise hand-drawn diagram components. This is made difficult by the presence of both geometric shapes and characters in diagrams. The goal of our research is to improve sketch recognition by improving the accuracy in grouping and classifying strokes in a diagram into text characters and shapes. We have done this by identifying the most significant features of strokes that can be used to distinguish shapes from text using a decision tree based partitioning technique. Implementation and evaluation of this new “shape divider” using these features against InkKit’s existing divider and the Microsoft divider has shown that our divider is more accurate at dividing text and shape strokes and can therefore improve overall sketch recognition.