人脸识别中基于均匀设计的SVM超参数调节方法

为了有效地解决超高维、多类别人脸识别中支持向量机(SVM,support vector machine)超参数调节计算复杂性问题,提出一种基于均匀设计(UD,uniform design)的SVM超参数调节的人脸识别方法。用UD代替传统的网格(grid)和梯度下降(gradient descent)方法挑选充分均衡分散在整个试验范围内且能得到满意实验结果的特征点,通过最小化k折交叉验证误差界或留一法(LOO,leave-one-out)误差界获取SVM最佳超参数。在UCI模拟数据集(Waveform)及人脸图像库(Yale,PIE)上进行了实验,结果表明,本文方法与传统的SVM超参数调节方法相比,能大大降低超参数调节的时间且能在一定程度上提高人脸分类识别率,使SVM超参数调节方法在解决高维人脸真实数据问题时具有一定的实用价值。

Uniform design based on SVM hyperparameters tuning for face recognition

To alleviate the computational complexity of support vector machine(SVM) hyperparameters tuning with high dimensionality in multiple face classification,a face recognition method is proposed based on SVM hyperparameters turning of uniform design(UD).The method replaces the grid and gradient descent method with UD to select some features that spread uniformly in the whole searching region and can gain satisfying testing results.The optimum SVM hyperparameters can be obtained by minimizing k-fold cross-valida...