Tahuti: A Sketch Recognition System for UML Class Diagrams

Bibliography Information:
Hammond, T., & Davis, R. (2002). Tahuti: A sketch recognition system for uml class diagrams. In AAAI Spring Symposium on Sketch Understanding (Vol. 3, No. 2).

URL:
http://rationale.csail.mit.edu/pubs/hammond/hammondsketchsymp2001.pdf

Tahuti is a sketch recognition environment designed around creating a "smarter" form of sketch recognition. The system is designed around recognizing shapes via geometric recognition, rather than requiring users draw a shape to match a template. This system uses a multi-layer framework for the recognition.Strokes are initially recognized as line segments, then grouped into more complex shapes if they are recognized as any particular shape.

The system recognizes seven "viewable" objects: a general class, an interface class, an inheritance
association, an aggregation association, a dependency association, an interface association, or a collection of unrecognized strokes. These objects can be further edited, which is a highly important feature in Tahuti that sets it apart from many other sketch recognition systems. For geometric recognition, Tahuti uses a variety of geometric "rules" that each shape follows. For instance, the way that Tahuti recognizes arrows is as follows:

1. Locate arrow shaft (identify points furthest away from each other. In the above picture, this would return the line segment between A and B)
2. Locate arrow heads (identify points furthest away from the shaft. This returns C and D)
3. Locate E, the point in the AB line that is twice the distance from B as there is distance between the line formed from the points CD to the arrowhead B.
4. Look to classify these line segments as either the arrow shaft, arrow head section, or "unclassified".
5. Based on the results from 4, classify arrow type as "dependency", "inheritance", "aggregation", or leave unclassified.

A preliminary study showed that the high accuracy of the system was pleasing to most users, who scored the system a 4.375/5 for drawing and a 4.825 for editing. The use experience was compared against an alternative paint program called Rational Rose, and most users preferred Tahuti for its ease of use, especially when it came to editing.

I believe that Tahuti was one of the early papers that influenced the Sketch Recognition Lab's focus on geometric recognition over template matching. Indeed, at this point in the field of sketch recognition, barely any of it is simply template matching. I also believe that it smartly uses the concepts of geometric properties from other sketch recognition techniques such as Rubine's features. This style of sketch recognition I believe is intuitive for testing applications in education in geometry, where the very systems used to recognize can be the ones used to grade basic geometric problems.