Nonnegative complementary prototype representation based classifier for object recognition

Based on the assumption that a query image can be represented as the nonnegative combination of the generated class-specific prototypes, a simple but effective classifier, non-negative complementary prototype representation based classifier (NCPRC), is proposed for object recognition. First, the query-dependent class prototypes are constructed using least squares. Second, we apply nonnegative least squares to estimate the coefficients of the prototypes. Finally, the identity of a query image is disclosed by the farthest rule with complementary prototypes. The proposed classifier doesn't need the delicate parameter tuning. Experiments on four standard image datasets validate the superiority of our approach over several state-of-the-art methods.