Volterra核优化的SRC人脸识别算法

Face Recognition Based on SRC with Volterra Kernels Optimization

为了提高稀疏表示分类算法对属于同一方向不同类别样本的分类准确率,提出了一种基于Volterra核优化的稀疏表示分类算法。该算法首先将原始的人脸图像分成不重叠的小块,并利用Volterra核映射到高维空间。在训练阶段遵循费舍尔标准,根据最大化类间距离和最小化类内距离来定义目标函数,从而获得优化Volterra核。与其他方法在ORL和YaleB标准数据集上进行对比实验,结果表明,采用Volterra核优化的SRC人脸识别分类方法在对样本的分类精度上提高了3%。

In order to improve the classification accuracy of sparse representation classification algorithm for samples belonging to the same direction but different categories,a sparse representation classification algorithm based on Volterra kernel optimization is proposed.Firstly,the original face images are divided into nonoverlapping blocks and mapped into high-dimensional space by Volterra kernel.In the training phase,Fisher criterion is followed,and the objective function is defined according to maximizing the distance between classes and minimizing the distance within classes to obtain the optimized Volterra kernel.The experimental results on ORL and YaleB standard datasets show that the classification accuracy of the SRC face recognition classification method based on Volterra kernel optimization is improved by 3%.