Abstract: Aiming at the problem that recognition rate of Principal Component Analysis(PCA) algorithm is low in face recognition, this paper proposes a modular PCA algorithm based on within-class weight average. Within-class weight average of each sub-image of all training samples in each class are calculated and used to normalize each corresponding sub-image of within-class sample. The best projection matrix from general matrix that is made up of all normalized sub-images is obtained accordingly. When all the sub-images of training samples and test samples are projected to the best projecting matrix got above, the recognition features is produced. The nearest distance classifier is used to distinguish each face. Experimental results on ORL face database indicate that the recognition performance of the algorithm is superior to that of general modular PCA algorithm.

Key words: face recognition, Principal Component Analysis(PCA), within-class weight average

摘要： 针对主成分分析(PCA)算法在人脸识别中识别率低的问题，提出一种基于类内加权平均值的模块PCA算法。该算法对每一类训练样本中每个训练样本的每个子块求类内加权平均值，用类内加权平均值对训练样本类内的相应子块进行规范化处理。由所有规范化后的子块构成总体散布矩阵，得到最优投影矩阵，由训练集全体子块的中间值对训练样本子块和测试样本子块进行规范化后投影到最优投影矩阵，得到识别特征，并用最近距离分类器分类。ORL人脸库上的实验结果表明，该算法的识别性能优于普通模块PCA算法。

关键词: 人脸识别, 主成分分析, 类内加权平均值

CLC Number: 

• TP391