Learning of a multi-class classifier with rejection option using sparse Representation

This paper introduces a multi-class classification algorithm based on sparse representation which considers on rejection option to minimize risks caused by outliers. Here the outliers include signals that do not belong to any classes learned in a training step. To successfully reject the outliers, new rejection measure and corresponding dictionary learning algorithm are presented. Experimental results on an image set, Caltech 101 [1] and one sound data set, AUI dataset, show that the proposed algorithm has improvements in classifying result by rejecting outliers.