Enhanced kernel estimation technique for pattern classification
K.P. Lam, E. Horne · Electronics Letters · 1993
The Letter reports the application of a nonparametric density estimation technique, the generalised K-nearest-neighbour (KNN) method, to a novel pattern classifier for binary images. In addition to offering an improved error rate performance over the fixed kernel method previously adopted, the method can be used to measure the inherent difficulty of a pattern classification problem because the nearest-neighbour error rate bounds the Bayes rate.