数学
嵌入
歧管(流体力学)
模式识别(心理学)
正定矩阵
判别式
人工智能
线性判别分析
组合数学
非线性降维
计算机科学
纯数学
特征(语言学)
算法
作者
Ying Hu,Benchao Li,Ruisheng Ran,Ji Feng
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-06-01
卷期号:: 1-7
标识
DOI:10.1109/ijcnn69854.2026.11729675
摘要
Symmetric positive definite (SPD) manifolds provide a fundamental representation for high-dimensional image sets and covariance descriptors by capturing second-order statistics and intrinsic geometric structures. However, in high-dimensional scenarios, SPD manifold data commonly suffer from the curse of dimensionality and geometric distortion, while conventional Euclidean dimensionality reduction methods fail to preserve Riemannian consistency. To address these challenges, this paper proposes a discriminative neighborhood preserving embedding on SPD manifolds (SPD-DNPE) with dual-mode neighborhood weighting strategies. By extending neighborhood preserving embedding(NPE) to the SPD Riemannian manifold and incorporating class label information, the proposed method enhances discriminability in the low-dimensional tangent space while preserving local manifold geometry. Extensive experiments on multiple benchmark datasets demonstrate that SPD-DNPE achieves superior classification accuracy and robustness compared with existing Grassmann and SPD manifold methods, kernel-based approaches, and several deep manifold models.
科研通智能强力驱动
Strongly Powered by AbleSci AI