【摘要】 针对人脸识别中单一特征难以取得理想效果的问题,提出了基于MBC和POEM特征融合的人脸识别方法.首先,在归一化的人脸图像上提取MBC编码图和POEM编码图,在每个编码图块上生成特征向量,应用线性判别分析对特征向量进行低维映射,并对其进行赋权相加得到最终相似度.所提算法在FERET的Dup1,Dup2,Fb和Fc 4个测试库上取得了较高的识别率,分别为93.77%,90.60%,99.58%和99.49%;在误识率为0.1%的条件下,在4个测试库上的认证率分别为95.70%,92.31%,99.75%和100%,进一步验证了该方法的有效性.更多还原

【Abstract】 Due to the representation difficulty of a face image by a single type feature used in face recognition,a MBC feature and POEMfeature-based face recognition scheme was proposed.Firstly,MBC and POEMcoding patterns were extracted from normalized face images. Secondly,feature vector of every block was generated by dividing the MBC and POEMcoding patterns into blocks. Finally,the classification capacity of features was enhanced by using weighted piecewise LDA algorithm. Recognition and verification test were carried out using the proposed algorithm on Dup1,Dup2,Fb and Fc,respectively,which were the four subsets of FERET. The recognition rates were 93. 77%,90. 60%,99. 58%,and 99. 49%,respectively,and the verification rates( false accepted rate is 0. 1%) were 95. 70%,92. 31%,99. 75%,and 100%,respectively. All these results indicated the effectiveness of the proposed algorithm.更多还原