The above paper1 provides an automated technique for decision tree design which relies only on a priori statistics. The technique is basically to find a partition of the class set which maximizes the probability of correct classification at a given decision node. To do so, in the above paper1 all the possible partitions are searched exhaustively. In the present comment it is pointed out that the search for the partitions with a size of two is sufficient, and the higher order of partitions need not be searched.

We propose a fast nearest neighbor finding algorithm, named tentatively an ordered partition, based on the ordered lists of the training samples of each projection axis. The ordered partition contains two properties, one is ordering¿to bound the search region, and the other is partitioning¿to reject the unwanted samples without actual distance computations. It is proved that the proposed algorithm can find k nearest neighbors in a constant expected time. Simulations show that the algorithm is rather distribution free, and only 4.6 distance calculations, on the average, were required to find a nearest neighbor among 10 000 samples drawn from a bivariate normal distribution.

In this letter we propose a method to represent the decision region in pattern recognition by a set of open parallelpipes within each of which some statistical property is satisfied. The proposed method has the following advantages: 1) it results in a great reduction of classification time, and 2) a priori knowledge can be easily incorporated in the decision rule.

A two-dimensional recursive estimation algorithm based on the asymmetric half-plane model is described for the problem of MMSE (minimum mean-square error) filtering. The optimum filtering problem is solved by formulating the asymmetric half-plane ARMA (autoregressive moving average) model for two-dimensional data. The sequential parameter identification from the noisy two-dimensional data is also discussed, utilizing the stochastic approximation. Experiments were performed for real image data, combining the proposed parameter identification and estimation algorithms. The results show that this method gives considerable improvement in SNR.