On-line sketch recognition for course of action diagrams

In this paper, we present a robust method of sketch recognition for course of action (COA) diagrams. User input is through free-hand sketching. COA symbols are recognized incrementally and the informal sketching input is replaced with formal vector graphs or images of the symbols. Multi-stroke COA symbols are divided into temporally continuous uni-stroke shapes. Firstly, it removes over-crossed end part of the uni-stroke shapes. Then, it extracts invariant geometric features of convex hull, largest-area inscribed and smallest-area enclosing polygons, perimeter and area ratios, and distinctive features of the concave and the open ended. Thirdly, the uni-stroke shapes are recognized by support vector machines (SVM). Fourthly, an intermediate feature recognizer is used to recognize shapes of intermediate complexity such as a dashed line, which improves sketch recognition performance in the COA domain. Finally, the system uses the LADDER shape definition language to represent the geometric properties of shapes, and is capable of recognizing common COA symbols of multi-stroke sketches and gesture commands with a recognition accuracy of more than 98%.