Joint Multiview Feature Learning for Hand-Print Recognition

In this article, we propose a joint multiview feature learning (JMvFL) method for hand-print recognition including both finger-knuckle-print (FKP) and palmprint recognition. Unlike most existing hand-print descriptors that are usually handcrafted and only focus on single-view features, our JMvFL method automatically and jointly learns multiview discriminant features of hand-print. Specifically, unlike the existing methods that extract features from raw pixels, we first form a multiview including both texture- and direction-view feature containers for hand-print images. Then, we aim to jointly learn multiview feature codes by enforcing three criteria: 1) the intraclass distance is minimized, and the interclass distance is maximized to make the feature codes of different classes more separate; 2) the information loss between the feature containers and the learned feature codes is minimized; and 3) the variance of interview feature codes is maximized so that the multiview feature codes are more complementary to enhance their overall discriminative power. Extensive experimental results demonstrate the effectiveness of the proposed method on various hand-print recognition tasks, including both FKP and palmprint recognition.