Deep hybrid manifold for image set classification

Xianhua Zeng, Jueqiu Guo, Yifan Wei, Yang Zhuo

Abstract: Highlights•The image set data is modeled as SPD manifold and Grassmann manifold data and used as input to the network simultaneously.•The Hybrid manifold autoencoder adds the reconstruction process and hybridizes complementary information of manifolds.•The Maximum fusion module fuses the classification results of each HMAE to prevent the information loss.