Complex matrix PCA consisting of two transform steps

摘要：

Recently developed matrix-based complex PCA (MCPCA) can fuse bimodal biometrics at the feature level and obtain a promising classification accuracy. However, when solving the eigen-equation, MCPCA is only a unsupervised method, which implies that in the training process MCPCA is not able to exploit the class labels of training samples. In order to overcome this problem, we propose an improvement to MCPCA, i.e. complex matrix PCA consisting of two transform steps. The proposed improvement has the following advantages: as the proposed improvement is a supervised method, it can fully exploit the class labels of training samples for training. Owing to the same reason, the solution of the proposed improvement can contain more discriminative information than MCPCA. The experimental results show that the proposed improvement to MCPCA can produce a good performance.

关键词： Face recognition Feature extraction Feature level fusion Biometrics MCPCA

作者: Ningbo Zhu Kaikai Lv Cong Li

作者单位: School of Computer and Communication Hunan University Changsha,China School of Computer and Communication Hunan University Changsha,China

会议类型: 国际会议

会议名称: 2010 International Conference on Future Information Technology(2010年未来信息技术国际会议 ICFIT 2010)

会议地点: 长沙

会议语种:英文

页码: 9-12

在线出版日期: 2010-12-14（万方平台首次上网日期，不代表论文的发表时间）