An Adaptive SVM Nearest Neighbor Classifier for Remotely Sensed Imagery

In this paper, we propose an extension of the kNN classifier based on the maximum margin principle. The proposed method is based on the idea to classify a given unlabeled sample by first finding its k nearest training samples. Then, a local partition of the feature space is carried out by means of local SVM decision boundaries determined after training a multiclass SVM classifier on the k training samples considered. The labeling of the unknown sample is done by looking at the local decision region it belongs to. The resulting global decision boundaries throughout the entire feature space are piecewise linear. The entire process can be however kernelized through the determination of the k nearest training samples in the kernel space by using a distance function simply reformulated on the basis of the adopted kernel. To illustrate the performance of the proposed method, an experimental analysis on different remote sensing data sets is reported and discussed.