The scaled probabilistic cleaning method is one of nonlinear noise reduction methods for chaotic time series, whose calculation quantity and needed memory volume increase exponentially with the number of reference points and the embedding dimension because of the joint processing of all data points in phase space. An optimized method, which modifies the estimation of forward probabilities and transition probabilities is proposed, and the computing workload is reduced to about 0.27 times that of the original method without degradation in noise reduction performance. The implementation of the method for the long time series also reduces the needed memory size.