Multiple Riemannian Manifold-Valued Descriptors Based Image Set Classification With Multi-Kernel Metric Learning

The importance of wild video based image set recognition is monotonically increasing due to the large amount data being collected by various devices including surveillance cameras, drive recorders, smart phones, and internet. content these videos often complex, it raises question how perform modeling feature extraction for set-based classification. In recent years, classification methods have advanced considerably in terms a covariance matrix, linear subspace, or Gaussian distribution. Moreover, distinctive geometry spanned them include Symmetric Positive Definite (SPD) manifold, Grassmannian embedded Riemannian respectively. As matter fact, most approaches just adopt single geometric model describe each given set, which may lose information useful To tackle this problem, we propose novel algorithm from multi-geometric perspective. Specifically, distribution are applied representation simultaneously. order fuse multiple heterogeneous manifold-valued features, well-equipped kernel functions first employed map into high dimensional Hilbert spaces. Then, multi-kernel metric learning framework devised embed learned hybrid kernels common lower subspace facilitate We conduct experiments on six widely used datasets representing different task: video-based face recognition, object categorization, emotion dynamic scene classification, cell identification, 3D hand pose estimation, evaluate performance proposed algorithm. extensive experimental results confirm its superiority over state-of-the-art methods.