In this paper a new technique is proposed to improve the recognition a
bility and the transaction speed to classify the Japanese and U.S. pap
er currency. Two types of data sets, time series data and Fourier powe
r spectra, are used in this study. In both cases, they are directly us
ed as inputs to the neural network. Still more we also refer a new eva
luation method of recognition ability. Meanwhile, a technique is propo
sed to reduce the input scale of the neural network without preventing
the growth of recognition. This technique uses only a subset of the o
riginal data set which is obtained using random masks. The recognition
ability of using large data set and a reduced data set are discussed.
In addition to that the results of using a reduced data set of the Fo
urier power spectra and the time series data are compared.