摘要
为进一步提高分块二维主成分分析(2DPCA)算法在人脸识别的识别率,提出一种人脸识别算法。将训练样本人脸矩阵按光照等相似条件进行分块并进行类内平均归一化;采用2DPCA算法构造特征空间,将分块矩阵在特征空间中进行投影得到训练样本识别特征,利用支持向量机(SVM)在分类上的优势,对训练样本识别特征和经过归一化分块2DPCA的测试样本识别特征进行分类,对人脸图像进行识别。选取ORL人脸数据库的图片进行实验,将该算法与传统2DPCA、2DPCA+SVM等算法进行比较,验证了该算法的性能优于其它算法。
To improve the face recognition rate of the algorithm of the modular two-dimensional principal component analysis （2DPCA）, a new face recognition algorithm was proposed. In the algorithm, the training images were divided into sub-images according to the illumination and they were normalized in each class. Through the algorithm of the 2DPCA, the feature space was obtained. The training images- recognition features were gotten by projecting the modular matrix to the feature space. Last- ly, support vector machine （SVM） was made full use of to classify the training images＇ recognition features and the test images＇ recognition features which were obtained by using the algorithm of the modular 2DPCA. Then the test images were recognized. Compared the proposed algorithm with the algorithm of the 2DPCA, the 2DPCA＋SVM etc. in the ORL face database, the results demonstrate that the performance of the proposed algorithm is better than that of the other algorithms.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第9期3229-3233,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61370180)
