Sub-manifolds sparsity preserving discriminant analysis
Jia He, Xiaobo Liu · 2025
In this paper, a novel method called sub-manifolds sparsity preserving discriminant analysis is proposed for the task of image set based face recognition. The proposed method /approach aims to learn the discriminative feature by integrating the relationship between hidden sub-manifolds in image set data. Firstly, each image set is modeled as a nonlinear manifold with a Gaussian mixture model comprising a number of Gaussian components, i.e., sub-manifolds, to handle the underlying local manifold structure. And then, it tries to preserve the sparse reconstructive relationship between these sub-manifolds when learning an embedding subspace in which the sample images within same sub-manifold and the sub-manifolds with same class label are compacted, and meanwhile the sub-manifolds with different class labels are separated. Since the sparse reconstructive relationship contains natural discriminative information, the proposed method can enhance its discriminative power for image set based face recognition. Experimental results show that the proposed method achieves better recognition performance and demonstrate its superiority over the state-of-the-art approaches.