Taking image as matrix, an eigenface algorithm uses eigenvalues and corresponding eigenvectors in recognition. The algorithm has advantage of no need of extracting geometric features of eyes, noses and mouths, but doesn't reach high recognition rate when single sample image per person is used for training. Another problem is that the larger the number of face modes is, the more complex the computation becomes. In this paper, an algorithm is proposed taking multi-samples as sub-modes and grouped face modes into small intersection ones to reduce computation and gain system extension property. In combination, the sum rule based on Bayesian theory is used. The face recognition experiment with the ORL and AR face databases shows that eigenface algorithm using multi-samples has reached a high recognition rate and a reasonable time cost. Applying distributed computation, the recognition system could be trained by grouped face modes and has the convenience of extension when new face modes are to be added.%本征脸法将图像看做矩阵，计算本征值和对应的本征向量作为代数特征进行识别，具有无需提取眼嘴鼻等几何特征的优点，但在单样本时识别率不高，且在人脸模式数较大时计算量大.将人脸模式的多个样本作为子模式，并将较多的人脸模式部分相交地分组，采用基于贝叶斯理论的和结合规则，以减小计算量和便于识别系统的扩展.采用ORL和AR图像库的实验表明，本征脸法在采用多样本训练后，识别率和识别时间都较好；识别系统可分布并行计算加快训练，在增加新人脸模式时，系统可以方便地进行扩展，并保持较高的识别率.
展开▼
