Nonnegative linear reconstruction measure based face recognition system

Nonnegative linear reconstruction measure (NLRM) can produce the sparse nonnegative representation coefficients which concentrate on the similar training samples. The samples with the same class-label are generally more similar than that with different class-labels. Based on this, we develop a NLRM based classifier (NLRMC). Using the decision rule of NLRMC, we continue to design a feature extractor, called nonnegative linear reconstruction projection (NLRP), such that NLRMC can achieve the optimal performance in the NLRP transformed subspace. NLRP combines with NLRMC to form the recognition system, which performs better than NLRP with other classifiers. Experiments are done on the Yale and AR face databases, and results demonstrate the recognition system of NLRP + NLRMC is more effective than other combinations.