shape properties (roughness, compactness, and elongation) did not represent the shape accurately not because of the use of LVQ. One reason we chose not to use LVQ is that it requires training. LVQ requires input of several shape features and determine the optimal classification result based on those features. The other and the main reason not to use LVQ is that TAC uses turn angle cross-correlation to better represent the shape and it cannot be input to LVQ.