Nonnegative discriminative encoded nearest points for image set classification

Image set classification has drawn much attention due to its rich set information. Recently, the most popular set-to-set distance-based representation methods have achieved interesting results by measuring the between-set distance. However, there are two intuitive assumptions, which are largely ignored: (1) The homogeneous samples should have positive contributions to approximate the nearest point in the probe set, while the heterogeneous samples should have no contributions and (2) the learned nearest points in each gallery set should have the lowest correlations. Therefore, this paper presents a novel method called nonnegative discriminative encoded nearest points for image set classification. Specifically, we use two explicit nonnegative constraints to ensure the coding coefficients sparse and discriminative simultaneously. Moreover, we additionally introduce a new class-wise discriminative term to further boost the discriminant ability of different sets. In this way, they can be boosted mutually so that the obtained coding coefficients are beneficial to the purpose of set classification. The results from extensive experiments and comparisons with some state-of-the-art methods on four challenging datasets demonstrate the superiority of our method.