One-against-one classification for zoom-endoscopy images

In this paper, we present a novel approach for the classification of zoom-endoscopy images based on the pit-pattern classification scheme. Our feature generation step is based on the computation of a set of statistical features in the wavelet-domain. In the classification step, we employ a one-against-one approach using 1-Nearest Neighbor classifiers together with sequential forward feature selection. Our experimental results show that this classification approach drastically increases leave-one-out crossvalidation accuracy for our six-class problem, compared to already existing approaches in this research area.