A Conditionally Positive Define Kernel Low-rank Subspace Clustering
Liping Chen · 2024
Abstract: Based on the view of feature sequence, this paper discusses the intrinsic mechanism of image sequence data kernel learning, whose original intention is to discover the nonlinear structural features in the high-dimensional feature space and the neighbourhood relationship of the original sample space. A Conditionally Positive Define Low Rank Kernel Subspace Clustering (Cpd - LKSC) is proposed to deal with the sequence relation of kernel subspace, which is used for clustering and dimensionality reduction. Because of the closure property of kernel function, we can combine different feature spaces by optimizing the learning of a self-expression low-rank kernel matrix, preserving the low dimensional structure of the mapping data in the feature space and the higher similarity between the vertices adjacent to the samples without training data . The conditional positive definite kernel of sequential self-expression is used to ensure the order feature. Results show Cpd - LKSC can obtain a low rank kernel structure that produces higher inter-class and lower intra-class variances than other algorithms.