Pattern recognition using average patterns of categorical k-nearest neighbors

The typical nonparametric method of pattern recognition "k-nearest neighbor rule (kNN)" is carried out by counting the labels of k-nearest training samples to a test sample. This method collects the k-nearest neighbors without taking into account a class, and it outputs the class of the test sample by using only the labels of neighborhoods. This work presents a classifier that outputs the class of a test sample by measuring the distance between the test sample and the average patterns, which are calculated using the k-nearest neighbors belonging to individual classes. A kernel method can be applied to this classifier for improving recognition rates. The performance of the proposed method is verified by experiments with benchmark data sets.