.png){kind=icon}
.png){kind=icon}
Return
Multi-layer double dictionary pair learning network via decoupling commonality and characteristic information
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-026-02287-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In recent years, dictionary pair learning has made remarkable achievements in the field of image classification. However, existing deep dictionary pair learning methods only focus on constructing synthetic and analytical dictionaries for each category, and rarely explore in depth the commonality and characteristic information between images of different categories, which limits their classification ability. In view of this, this paper innovatively proposes a Multi-layer Double Dictionary Pair Learning Network (MDDPL-Net) via decoupling commonality and characteristic information for image classification. Specifically, by jointly constructing multiple class-specific sub-dictionary pairs and a class-shared dictionary pair, the proposed method can capture the most discriminative features of each category and the shared features between different categories of images respectively. Besides, to enhance the sparsity of feature representation and improve its discriminative ability, we impose sparsity constraint on the class-specific dictionaries and low-rank constraint on the class-shared dictionary. Furthermore, we achieve different levels of representation learning by stacking multiple double dictionary pair layers. Numerous experimental results indicate that the proposed method achieves significant improvement in image classification tasks.
Keywords:
Deep dictionary learning
Dictionary pair learning
Class-specific information
Class-shared information
Feature decoupling