Abstract: This paper proposes a novel descriptor EVP（Edge Vector Pattern）, and the combination of common classifiers is composed for face recognition, to compare with other known methods. EVP is characterized not only by using gradient values instead of the pixel intensity values to describe local object appearance and shape, but also describes the relationship between neighborhood information. Based on the FERET face image database, when using the nearest neighbor classifier, EVP is superior to other methods. Based on the LFW database, and the combination of MKL-SVM, the result is impressive. More commendable is that EVP is roughly 20 times faster than methods based on Gabor, while Gabor method is now considered as one of the best methods.

Key words: Edge Vector Pattern（EVP）, Multiple Kernel Learning-Support Vector Machine（MKL-SVM）, Gabor, Local Binary Pattern（LBP）, real time face recognition

摘要： 提出了新的EVP（边缘向量模式）人脸描述（或表示）方法，并结合常见分类器组成人脸识别系统，与其他已有方法比较。EVP的特点是不仅用梯度值代替像素强度值描述对象的外观和形状，还描述了邻近区域信息之间的关系。对于FERET人脸数据库的实验结果显示，当结合最近邻分类器时，EVP优于其他表示方法。对于LFW数据库，并结合多核支持向量机（MKL-SVM）的EVP，得到的结果非常令人满意。更可贵的是，基于EVP方法的识别速度大约是基于伽柏（Gabor）方法的20倍。而基于伽柏的方法是目前公认较好的方法。

关键词: 边缘向量模式（EVP）, 多核支持向量机, 伽柏, 局部二值模式（LBP）, 实时人脸识别