关键词: 掌纹识别, 特征提取, 双向主成分分析（BDPCA）, 散度矩阵

Abstract: Under the condition of small sample size, the average of all training samples used in the Bi-Directional PCA algorithm is the scatter center of the samples. This algorithm can not guarantee the optimality of the eigenvalues. In order to solve this problem, this paper proposes an improved BDPCA palmprint identification algorithm which is based on sample scatter matrix. To reconstruct the overall scatter matrix, the algorithm adopts the K-values matrix of the training samples instead of the average matrix of the training samples. The algorithm is tested using PolyU and CASIA. The results show that the improved method is more optimal in recognition rate than the traditional BDPCA.

Key words: palmprint recognition, feature extraction, Bi-Directional Principal Component Analysis（BDPCA）, scatter matrix