LMDAPNet: A Novel Manifold-Based Deep Learning Network

In this paper, we propose a novel deep learning network, called Local Manifold Discriminant Analysis Projection Network (LMDAPNet). Different from most existing face recognition based on Deep Neural Networks (DNNs) methods that learn complex feature representations of data itself, our method aims to exploit the discriminant and geometrical structure of data manifold by optimally preserving the local neighborhood information. Firstly, it learns the local discriminant embedding for the submanifold of each class by solving an optimization problem. Secondly, we propose a Local Manifold Discriminant Analysis Projection (LMDAP) algorithm to learn convolutional kernel based on the local discriminant embedding space. Meanwhile, we define a new subspace-to-subspace distance metric to measure the dissimilarity between manifold pair. Finally, the learned convolutional kernel is used to extract feature maps efficiently. After a series of feature extraction, multiscale feature analysis is to encode the feature maps before feeding into a classifier. Experimental results on four benchmark datasets show that our proposed network is more robust and competitive performance for face recognition.

Abstract licence: CC BY-NC-ND

42 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.