利用异或运算和编码约束的降维LDP人脸识别方法

Face Recognition Method Using Reduced-dimension LDP Using XOR and Coding Constraints

摘要:

为了有效表示面部特征,在局部方向模式(LDP)的基础上,提出降维局部方向模式(RDLDP)；首先,修改LDP编码模式约束以完成模式的重构,通过对LDP码进行异或运算来计算每个块的单一码；然后,将所得编码图像划分为生成直方图,连接所有区域的直方图块以形成最终描述符;最后,计算特征向量间的卡方相异性度量值,并使用最近邻分类器完成最终的人脸识别;实验采用了3个公开的标准数据库FERET、扩展YALE-B和ORL,提出的改进方法在3个数据集上的最高识别率分别可高达96.97%、96.10%、97.61%,该结果验证了提出方法的有效性。与其他基于局部描述符的先进方法相比,提出方法在准确度和错误识别率等方面更优。

Abstract:

To represent facial features effectively, on the basis of local directional patterns (LDP), a reduced -dimension-local directional pattern (RDLDP) is proposed. Firstly, the constraints of LDP encoding mode is modified to complete the pattern reconstruction, and through the XOR of the LDP code, code of each block is calculated. Then, the encoding image is divided into histograms, and the histograms of all areas are connected to form the final descriptor. Finally, the chi square dissimilarity measure between the eigenvectors is computed, and the k-nearest neighbor classifier is adopted to complete the final face recognition. Three public available standard databases, FERET, extended YALE-B, and ORL are adopted in the experiment. The proposed method can be up to 96.97%, 96.10% and 97.61% respectively in the three data sets.And the effectiveness of the proposed algorithm verified by experimental results. Compared with other advanced methods based on local descriptors, the proposed method is superior in accuracy and error recognition rate.