This paper investigates the actual situation of training of Fuzzy Spline Curve identifier (FSCI). FSCI is a primitive curve identification system which has been proposed to es tablishe a general-purpose freehand interface for computer aided drawing (CAD) systems. It succeeded in distinguishing a freehand drawing into seven kinds of primitive curves which are indispensable for CAD. The key was the introduction of the fuzzy reasoning which embodied a strategy to try to find the simplest primitive curves based on user's drawing manner. Furthermore, a trainable version of FSCI has learning ability to adjust its fuzzy reasoning rules (which are materialized a a fuzzy neural network) and adapt itself to individual user's drawing manner. This paper sets up some experiment to train this FSCI, and demonstrates how the adjustement of the fuzzy neural network improves FSCI's curve class recognition rates. Then, through some detailed observations on a concerete tranining result from one user, it shows how the adjusted fuzzy neural network explicitly explains the effect of the training.
展开▼