Full Length Research Paper
Abstract
Single labeled biometric recognition is one of the main challenges to graph-based transductive classification learning. To enhance the recognition rate of single labeled problem, sparse representation provides a feasible strategy for representation learning. In this paper, we developed a power l1-graph learning technique for semi-supervised learning, called label propagation by power l1-graph (LPPG). Different from all existing graph-based methods, we assume that the similarity relationship in the label space is a power function in the sample space. What is important is that the determinated power value measured by sparseness is given. Our method characterizes the second sparse processing, and seeks to find a reasonable label propagation way. This characteristic makes our algorithm more intuitive and more powerful than those methods based on the original l1-graph. This proposed method is applied to biometrics recognition and the experiment results show that our algorithm consistently outperforms those original l1-graph-based methods. This demonstrates that our method is a good choice for real-world biometrics applications, especially when there is only one labeled image.
Key words: Power function, sparse reconstruction, label propagation, biometrics recognition.
Copyright © 2025 Author(s) retain the copyright of this article.
This article is published under the terms of the Creative Commons Attribution License 4.0