基于核空间Fisher鉴别字典稀疏表示的图像分类

Sparse coding is popular for image classification and its performance is the state-of-the art. However, these methods design linear classifiers for the purpose of classifying when performing sparse coding. The performance of these methods will degrade if the features are linear inseparable. Thus, a novel image classification method is proposed in this paper. The proposed method is based on the sparse coding of Fisher discrimination dictionary in kernel space. First, the input image patches are mapped into high dimension feature space, and then, the sparse coding are performed in the feature space via introducing the Fisher discrimination criterion. The proposed method can not only improve the discriminative ability of spare dictionary but also overcome the shortcoming of existing methods which can not deal with the linear inseparable case. Experimental results verify the effectiveness of the proposed method.