Learning Vector Quantization (LVQ) has been intensively studied to gen
erate good reference models in pattern recognition since 1986, and it
has some nice theoretical properties. However, the design of reference
models based on LVQ suffers from several major drawbacks for the reco
gnition of large-set patterns, in which good reference models play an
important role in achieving high performance. They are due in large pa
rt to the following facts. (1) it may not generate good reference mode
ls, if the initial values of the reference models are outside the conv
ex hull of the input data, (2) it cannot guarantee optimal reference m
odels due to the strategy to accept new reference models in each itera
tion step, and (3) it is apt to get stuck at overtraining phenomenon.
In this paper, we first discuss the impact of these problems. And then
, to cope with these, we propose a new method for the optimal design o
f large-set reference models using an improved LVQ3 combined with Simu
lated Annealing which has been proven to be a useful technique in many
areas of optimization problems. Experimental results with large-set h
andwritten characters reveal that the proposed method is superior to t
he conventional method based on averaging and other LVQ-based methods.