一种多算法融合的人脸识别方法研究

针对人脸识别技术易受光照、姿态、表情等影响 ,为了增强人脸识别算法的鲁棒性,提出首先采用 LBP算法提取人脸图像的局部纹理特征,使用PCA算法将高维的空间人脸图像投影到低维的 特征空间,使 用LDA算法利用人脸类别标签信息寻找最优的投影向量,实现了人脸图像维度进一步地压缩 ,最后使用SVM 分类器分类匹配得到识别结果。分别使用ORL和Yale人脸数据库验证了算法的有效性,实 验结果表明,文 中该方法具有良好的识别性能,与其它的识别算法相比,识别率有了较大的提高。

Research on a face recognition method based on multi-algorithms fusion

Face recognition technology was vulnerable to illumination,posture,e xpression and other factors,in order to enhance the robustness of face recognition algorithm,LBP algorithm was used to extract local texture features of face images first,then PCA algorithm was used to project high-dimen sional spatial face images into low dimensional feature space,using LDA algorithm to find the optimal projection ve ctors with face class label information,the dimension of image was further compressed.Finally,compared wi th other classical classifiers such as the K-Nearest Neighbor classifier,Bayes classifier,SVM classifier was used to get great recognition and classification results.ORL and Yale face databases were used to verify the effe ctiveness of the algorithm.In ORL, the recognition rate of the method PCA gets 77% only,PCA combined with LBP gets 83.0%,PCA combined with LDA gets 91.5%,the algorithm SRC based on sparse representation and dictionary learning gets 91.5%,and the proposed method gets 100%,which has been greatly improved compared with other r ecognition algorithms.In Yale,the recognition rate of the method PCA gets 74% only,PCA combined with LB P gets 87.9%,PCA combined with LDA gets 93.3%,the algorithm SRC based on sparse representation and dictio nary learning gets 85.6%,and the proposed method gets 100%.The testing results illustrate that the proposed method has also great recognition performance in this database,which shows this method has feasibility and effectiveness.