In performing iconic gestures, we do not precisely draw the
shape of the object, but depict the salient characteristics of the
object’s shape. Based on this assumption, we propose a
method of assigning gesture shapes by classifying the objects
into four basic shapes: circle, rectangle, linear, and other
shapes. For this purpose, we employ simple image processing,
clustering, and machine learning techniques. Fig. 1 shows the
flow of our method.