噪声条件下对稀疏表示分类器的优化

Optimization of Sparse Representation Classifier Under Noisy Conditions

传统的稀疏表示是直接使用所有的训练样本来线性表示测试样本,它依赖以下核心思想:即测试样本可以由相同类别的少量训练样本来表示。然而,当学习给定字典的稀疏表示时,还不能很好地解决不同的面部表情、姿势,和不同照明条件的人脸识别问题。这些噪声会极大地影响表示的准确度。因此,探索一种更有效的方法来表示测试样本是一个至关重要的问题。提出引入局部线性编码的方案并且融合样本之间的空间距离信息,来优化在噪声条件下稀疏表示分类器的分类精度。

The traditional sparse representation is to directly use all the training samples to represent the test samples linearly. It relies on the following core idea： test samples can be represented by a small number of training samples in the same category. However, when learning the sparse representation of a given dictionary. Different facial expressions, postures, and face recognition problems under different lighting conditions are not well solved. These noises can greatly affect the accuracy of the representation. Therefore, explores a more effective way to represent test samples is a crucial issue. Proposes a scheme of introducing local linear coding and the spatial distance information between samples is fused to optimize the classification accuracy of classifiers sparsely under noisy conditions.